Commentary: Sure, use Meta’s Muse bot to help with your chores — if you want Mark Zuckerberg to know all your secrets

If Silicon Valley has perfected any business model, it’s the one in which it cooks up a new technology and then persuades people they can’t live without it. The newest example is Meta Platforms’ personal artificial intelligence agent, Muse.

Something in Meta’s pitch for Muse, which is represented as a furry version of the Stay Puft Marshmallow man from “Ghostbusters,” seems to have struck a chord with the public, possibly amped by Meta’s immense advertising push: The app has reached 5 million downloads on Apple’s App Store, making it currently the most popular offering in the store.

Meta makes enticing claims for Muse’s capabilities. The company boasts that Muse can read and write answers to your emails. One of its TV commercials has it sorting the user’s emails to prioritize “what’s important,” placing a carpool schedule on the family calendar, ordering school supplies and turning on her stereo for a house party. The user is shown devoting her free time to chopping vegetables for said party.

Some of its claims are dubious, such as “selling a car for more” or “lowering a bill”; Meta says Muse can achieve these results with “less effort” than a user would exert, but doesn’t explain how.

And some are disquieting. “Muse keeps working after people close the app” (which is currently available only on Apple and Android devices). Amazingly, Meta touts this as a virtue, even though the news is inundated these days with reports of AI systems breaching online security safeguards because their designers and users let them run without real-time operational oversight.

An AI bot that goes on its merry way and informs the user what it’s been up to only when it chooses to? Not the key to a happy relationship.

Meta isn’t alone in trying to convince ordinary people that AI is indispensable. AI companies are encouraging people to upload their health information to be scrutinized by their bots, or to grant access to their bank or investment accounts. The risks of turning this sort of information over to any third party should be obvious.

Meta touts Muse as a free service, but that’s misleading. The company says it expects to make a profit by charging fees for Muse transactions — but as consumers have learned from the proliferation of retailer fees for credit card purchases, those charges can end up costing the consumers themselves. Meta also says Muse is free only up to a certain level of usage, after which users will have to buy a subscription for up to $100 a month. I asked Meta how much usage would trigger the subscription requirement, but it didn’t give me an answer.

Two major issues have already surfaced among Muse’s early users. One is why anyone living a normal life needs a bot to perform such quotidian tasks as reading and writing emails, placing appointments on a calendar or calling a plumber.

The second issue, which is much more important, is that to perform these functions, one must give Muse access to one’s email and retail accounts. Let’s start with that. The opinion of privacy advocates who have weighed in on this requirement generally boils down to: Anyone doing this should have their head examined.

It’s not merely that security breaches exposing the personal information of up to hundreds of millions of people have become almost daily occurrences. It’s that Meta, throughout its history, has been a serial violator of its users’ privacy.

Things came to a head in 2019, when the Federal Trade Commission imposed a record $5-billion fine on the company to settle charges that it had been “deceiving users about their ability to control the privacy of their personal information.” What was striking about that case was that the FTC brought it because it caught the company, then known as Facebook, violating a 2012 consent order that involved similar accusations. As big as the fine appeared, at the time, Facebook collected as much in revenue every month; last year, it collected that much every nine days.

Meta’s misuse of its users’ information was addressed in the $17-billion payment it agreed to in August to settle accusations from California and other states that it knew that features of Facebook and Instagram exposed children using those services to bullies, extortionists and sexual predators.

In the words of California Atty. Gen. Rob Bonta, the company “designed and deployed harmful features on Instagram and Facebook that drive compulsive use by children and teens to their mental and physical detriment, all the while misleading users, their families, and the public regarding the existence and severity of these risks.” (The $17-billion settlement is equivalent to Meta’s revenue for a single month in 2025.)

The company didn’t formally admit to the charges in these cases, but one would have to be hopelessly naive to think that means it’s innocent.

Put all that together and “there’s nothing about Meta’s track record that builds anything near trust that they are going to be responsible stewards or custodians of personal information,” said Lee Hepner, senior legal counsel at the American Economic Liberties Project. “I can’t think of a company with a deeper record of betraying the public trust.”

The defense given by Mark Zuckerberg, the company’s chief executive, chairman and co-founder, of his company’s historical nonchalance about privacy violations has been a sort of shrugging acknowledgment that this is what happens when your focus is on making users’ world “a little better.” In the past, he has copped to his wrongdoing and promised not to do it again, pleading that “we really messed this one up,” “we simply did a bad job with this release,” or “sometimes we move too fast.”

Meta wants us to think that it has learned its lesson. In response to my question about its record on privacy, a company spokesperson referred me to a company post asserting that its Muse agent is endowed with “first-of-its-kind privacy, safety, and security protections … that no other agent provides,” and that “Muse has no visibility into people’s passwords or payment methods. … Muse can use them without seeing them.”

Unexpected leakages of private information already have been cited by early Muse users. Tech commentator Matt Robb reported that Muse gave a would-be buyer of an item he posted for sale a price below his specification and gave the buyer his home address. (The Muse team attributed the glitch to confusing settings in Robb’s Muse account.)

Jason Aten, a tech columnist at Inc. magazine, reported that his Muse read and regurgitated his private messages even though he believed he had denied it that capability. Muse engineers said the bot had exploited a permissions setting on his Mac computer he didn’t realize was applicable.

Those cases point to complexities in the Muse settings that confound experts, let alone laypersons. That undermines Meta’s claim that Muse is “simple to use. People just tell Muse what needs to get done, and it takes action.”

That leaves us with the question of why anyone needs Muse. To answer that question for myself, I launched the sign-up process via my Facebook account, but since the first thing it asked (after prompting me to give the little creature a name) was access to my Gmail account, I shut it down. But other fairly savvy users among tech mavens don’t point to urgent needs that only Muse can satisfy.

Allison Johnson of the Verge writes that her bot successfully emailed a junkman to remove a fallen tree branch in her yard, helped her order camping equipment and put a part for her espresso machine in her online cart but couldn’t complete the purchase. New York Times writer Eli Tan used it to call his dental insurer and wait on hold for a human representative to come on. It also arranged his credit card statements in a spreadsheet and helped him pursue his hobby of collecting ceramic bowls, a task which strikes me as having a fairly small community of potential users.

Muse informed Ryan Broderick, the proprietor of tech blog Garbage Day, that “it could shop for me, book dinner reservations, or … even do tasks on my laptop. But I didn’t need to do any of those things yesterday. And I definitely didn’t need to do any of them so badly that I was willing to risk it deleting something important, hallucinating an email to someone, or buying me something I didn’t want.”

That’s an understandable take. None of the chores that Meta boasts about are beyond the capabilities of most people living in today’s digital world. Booking a dinner reservation? Buying something online? (As it happens, Amazon has blocked Muse purchases on its site for the moment, saying that the access violates its terms of service; Meta says Muse can make purchases on other online sites such as Walmart and Expedia.)

Everyone will have to make their own decision about whether the services Muse provides are worth the risks that Broderick cited. Nor should anyone be fooled by the smiley-faced icon Meta chose to represent Muse. For all anyone knows, it’s a wolf in furry clothing.

If Silicon Valley has perfected any business model, it’s the one in which it cooks up a new technology and then persuades people they can’t live without it. The newest example is Meta Platforms’ personal artificial intelligence agent, Muse.

Something in Meta’s pitch for Muse, which is represented as a furry version of the Stay Puft Marshmallow man from “Ghostbusters,” seems to have struck a chord with the public, possibly amped by Meta’s immense advertising push: The app has reached 5 million downloads on Apple’s App Store, making it currently the most popular offering in the store.

Meta makes enticing claims for Muse’s capabilities. The company boasts that Muse can read and write answers to your emails. One of its TV commercials has it sorting the user’s emails to prioritize “what’s important,” placing a carpool schedule on the family calendar, ordering school supplies and turning on her stereo for a house party. The user is shown devoting her free time to chopping vegetables for said party.

Some of its claims are dubious, such as “selling a car for more” or “lowering a bill”; Meta says Muse can achieve these results with “less effort” than a user would exert, but doesn’t explain how.

And some are disquieting. “Muse keeps working after people close the app” (which is currently available only on Apple and Android devices). Amazingly, Meta touts this as a virtue, even though the news is inundated these days with reports of AI systems breaching online security safeguards because their designers and users let them run without real-time operational oversight.

An AI bot that goes on its merry way and informs the user what it’s been up to only when it chooses to? Not the key to a happy relationship.

Meta isn’t alone in trying to convince ordinary people that AI is indispensable. AI companies are encouraging people to upload their health information to be scrutinized by their bots, or to grant access to their bank or investment accounts. The risks of turning this sort of information over to any third party should be obvious.

Meta touts Muse as a free service, but that’s misleading. The company says it expects to make a profit by charging fees for Muse transactions — but as consumers have learned from the proliferation of retailer fees for credit card purchases, those charges can end up costing the consumers themselves. Meta also says Muse is free only up to a certain level of usage, after which users will have to buy a subscription for up to $100 a month. I asked Meta how much usage would trigger the subscription requirement, but it didn’t give me an answer.

Two major issues have already surfaced among Muse’s early users. One is why anyone living a normal life needs a bot to perform such quotidian tasks as reading and writing emails, placing appointments on a calendar or calling a plumber.

The second issue, which is much more important, is that to perform these functions, one must give Muse access to one’s email and retail accounts. Let’s start with that. The opinion of privacy advocates who have weighed in on this requirement generally boils down to: Anyone doing this should have their head examined.

It’s not merely that security breaches exposing the personal information of up to hundreds of millions of people have become almost daily occurrences. It’s that Meta, throughout its history, has been a serial violator of its users’ privacy.

Things came to a head in 2019, when the Federal Trade Commission imposed a record $5-billion fine on the company to settle charges that it had been “deceiving users about their ability to control the privacy of their personal information.” What was striking about that case was that the FTC brought it because it caught the company, then known as Facebook, violating a 2012 consent order that involved similar accusations. As big as the fine appeared, at the time, Facebook collected as much in revenue every month; last year, it collected that much every nine days.

Meta’s misuse of its users’ information was addressed in the $17-billion payment it agreed to in August to settle accusations from California and other states that it knew that features of Facebook and Instagram exposed children using those services to bullies, extortionists and sexual predators.

In the words of California Atty. Gen. Rob Bonta, the company “designed and deployed harmful features on Instagram and Facebook that drive compulsive use by children and teens to their mental and physical detriment, all the while misleading users, their families, and the public regarding the existence and severity of these risks.” (The $17-billion settlement is equivalent to Meta’s revenue for a single month in 2025.)

The company didn’t formally admit to the charges in these cases, but one would have to be hopelessly naive to think that means it’s innocent.

Put all that together and “there’s nothing about Meta’s track record that builds anything near trust that they are going to be responsible stewards or custodians of personal information,” said Lee Hepner, senior legal counsel at the American Economic Liberties Project. “I can’t think of a company with a deeper record of betraying the public trust.”

The defense given by Mark Zuckerberg, the company’s chief executive, chairman and co-founder, of his company’s historical nonchalance about privacy violations has been a sort of shrugging acknowledgment that this is what happens when your focus is on making users’ world “a little better.” In the past, he has copped to his wrongdoing and promised not to do it again, pleading that “we really messed this one up,” “we simply did a bad job with this release,” or “sometimes we move too fast.”

Meta wants us to think that it has learned its lesson. In response to my question about its record on privacy, a company spokesperson referred me to a company post asserting that its Muse agent is endowed with “first-of-its-kind privacy, safety, and security protections … that no other agent provides,” and that “Muse has no visibility into people’s passwords or payment methods. … Muse can use them without seeing them.”

Unexpected leakages of private information already have been cited by early Muse users. Tech commentator Matt Robb reported that Muse gave a would-be buyer of an item he posted for sale a price below his specification and gave the buyer his home address. (The Muse team attributed the glitch to confusing settings in Robb’s Muse account.)

Jason Aten, a tech columnist at Inc. magazine, reported that his Muse read and regurgitated his private messages even though he believed he had denied it that capability. Muse engineers said the bot had exploited a permissions setting on his Mac computer he didn’t realize was applicable.

Those cases point to complexities in the Muse settings that confound experts, let alone laypersons. That undermines Meta’s claim that Muse is “simple to use. People just tell Muse what needs to get done, and it takes action.”

That leaves us with the question of why anyone needs Muse. To answer that question for myself, I launched the sign-up process via my Facebook account, but since the first thing it asked (after prompting me to give the little creature a name) was access to my Gmail account, I shut it down. But other fairly savvy users among tech mavens don’t point to urgent needs that only Muse can satisfy.

Allison Johnson of the Verge writes that her bot successfully emailed a junkman to remove a fallen tree branch in her yard, helped her order camping equipment and put a part for her espresso machine in her online cart but couldn’t complete the purchase. New York Times writer Eli Tan used it to call his dental insurer and wait on hold for a human representative to come on. It also arranged his credit card statements in a spreadsheet and helped him pursue his hobby of collecting ceramic bowls, a task which strikes me as having a fairly small community of potential users.

Muse informed Ryan Broderick, the proprietor of tech blog Garbage Day, that “it could shop for me, book dinner reservations, or … even do tasks on my laptop. But I didn’t need to do any of those things yesterday. And I definitely didn’t need to do any of them so badly that I was willing to risk it deleting something important, hallucinating an email to someone, or buying me something I didn’t want.”

That’s an understandable take. None of the chores that Meta boasts about are beyond the capabilities of most people living in today’s digital world. Booking a dinner reservation? Buying something online? (As it happens, Amazon has blocked Muse purchases on its site for the moment, saying that the access violates its terms of service; Meta says Muse can make purchases on other online sites such as Walmart and Expedia.)

Everyone will have to make their own decision about whether the services Muse provides are worth the risks that Broderick cited. Nor should anyone be fooled by the smiley-faced icon Meta chose to represent Muse. For all anyone knows, it’s a wolf in furry clothing.

If Silicon Valley has perfected any business model, it’s the one in which it cooks up a new technology and then persuades people they can’t live without it. The newest example is Meta Platforms’ personal artificial intelligence agent, Muse.

Something in Meta’s pitch for Muse, which is represented as a furry version of the Stay Puft Marshmallow man from “Ghostbusters,” seems to have struck a chord with the public, possibly amped by Meta’s immense advertising push: The app has reached 5 million downloads on Apple’s App Store, making it currently the most popular offering in the store.

Meta makes enticing claims for Muse’s capabilities. The company boasts that Muse can read and write answers to your emails. One of its TV commercials has it sorting the user’s emails to prioritize “what’s important,” placing a carpool schedule on the family calendar, ordering school supplies and turning on her stereo for a house party. The user is shown devoting her free time to chopping vegetables for said party.

Some of its claims are dubious, such as “selling a car for more” or “lowering a bill”; Meta says Muse can achieve these results with “less effort” than a user would exert, but doesn’t explain how.

And some are disquieting. “Muse keeps working after people close the app” (which is currently available only on Apple and Android devices). Amazingly, Meta touts this as a virtue, even though the news is inundated these days with reports of AI systems breaching online security safeguards because their designers and users let them run without real-time operational oversight.

An AI bot that goes on its merry way and informs the user what it’s been up to only when it chooses to? Not the key to a happy relationship.

Meta isn’t alone in trying to convince ordinary people that AI is indispensable. AI companies are encouraging people to upload their health information to be scrutinized by their bots, or to grant access to their bank or investment accounts. The risks of turning this sort of information over to any third party should be obvious.

Meta touts Muse as a free service, but that’s misleading. The company says it expects to make a profit by charging fees for Muse transactions — but as consumers have learned from the proliferation of retailer fees for credit card purchases, those charges can end up costing the consumers themselves. Meta also says Muse is free only up to a certain level of usage, after which users will have to buy a subscription for up to $100 a month. I asked Meta how much usage would trigger the subscription requirement, but it didn’t give me an answer.

Two major issues have already surfaced among Muse’s early users. One is why anyone living a normal life needs a bot to perform such quotidian tasks as reading and writing emails, placing appointments on a calendar or calling a plumber.

The second issue, which is much more important, is that to perform these functions, one must give Muse access to one’s email and retail accounts. Let’s start with that. The opinion of privacy advocates who have weighed in on this requirement generally boils down to: Anyone doing this should have their head examined.

It’s not merely that security breaches exposing the personal information of up to hundreds of millions of people have become almost daily occurrences. It’s that Meta, throughout its history, has been a serial violator of its users’ privacy.

Things came to a head in 2019, when the Federal Trade Commission imposed a record $5-billion fine on the company to settle charges that it had been “deceiving users about their ability to control the privacy of their personal information.” What was striking about that case was that the FTC brought it because it caught the company, then known as Facebook, violating a 2012 consent order that involved similar accusations. As big as the fine appeared, at the time, Facebook collected as much in revenue every month; last year, it collected that much every nine days.

Meta’s misuse of its users’ information was addressed in the $17-billion payment it agreed to in August to settle accusations from California and other states that it knew that features of Facebook and Instagram exposed children using those services to bullies, extortionists and sexual predators.

In the words of California Atty. Gen. Rob Bonta, the company “designed and deployed harmful features on Instagram and Facebook that drive compulsive use by children and teens to their mental and physical detriment, all the while misleading users, their families, and the public regarding the existence and severity of these risks.” (The $17-billion settlement is equivalent to Meta’s revenue for a single month in 2025.)

The company didn’t formally admit to the charges in these cases, but one would have to be hopelessly naive to think that means it’s innocent.

Put all that together and “there’s nothing about Meta’s track record that builds anything near trust that they are going to be responsible stewards or custodians of personal information,” said Lee Hepner, senior legal counsel at the American Economic Liberties Project. “I can’t think of a company with a deeper record of betraying the public trust.”

The defense given by Mark Zuckerberg, the company’s chief executive, chairman and co-founder, of his company’s historical nonchalance about privacy violations has been a sort of shrugging acknowledgment that this is what happens when your focus is on making users’ world “a little better.” In the past, he has copped to his wrongdoing and promised not to do it again, pleading that “we really messed this one up,” “we simply did a bad job with this release,” or “sometimes we move too fast.”

Meta wants us to think that it has learned its lesson. In response to my question about its record on privacy, a company spokesperson referred me to a company post asserting that its Muse agent is endowed with “first-of-its-kind privacy, safety, and security protections … that no other agent provides,” and that “Muse has no visibility into people’s passwords or payment methods. … Muse can use them without seeing them.”

Unexpected leakages of private information already have been cited by early Muse users. Tech commentator Matt Robb reported that Muse gave a would-be buyer of an item he posted for sale a price below his specification and gave the buyer his home address. (The Muse team attributed the glitch to confusing settings in Robb’s Muse account.)

Jason Aten, a tech columnist at Inc. magazine, reported that his Muse read and regurgitated his private messages even though he believed he had denied it that capability. Muse engineers said the bot had exploited a permissions setting on his Mac computer he didn’t realize was applicable.

Those cases point to complexities in the Muse settings that confound experts, let alone laypersons. That undermines Meta’s claim that Muse is “simple to use. People just tell Muse what needs to get done, and it takes action.”

That leaves us with the question of why anyone needs Muse. To answer that question for myself, I launched the sign-up process via my Facebook account, but since the first thing it asked (after prompting me to give the little creature a name) was access to my Gmail account, I shut it down. But other fairly savvy users among tech mavens don’t point to urgent needs that only Muse can satisfy.

Allison Johnson of the Verge writes that her bot successfully emailed a junkman to remove a fallen tree branch in her yard, helped her order camping equipment and put a part for her espresso machine in her online cart but couldn’t complete the purchase. New York Times writer Eli Tan used it to call his dental insurer and wait on hold for a human representative to come on. It also arranged his credit card statements in a spreadsheet and helped him pursue his hobby of collecting ceramic bowls, a task which strikes me as having a fairly small community of potential users.

Muse informed Ryan Broderick, the proprietor of tech blog Garbage Day, that “it could shop for me, book dinner reservations, or … even do tasks on my laptop. But I didn’t need to do any of those things yesterday. And I definitely didn’t need to do any of them so badly that I was willing to risk it deleting something important, hallucinating an email to someone, or buying me something I didn’t want.”

That’s an understandable take. None of the chores that Meta boasts about are beyond the capabilities of most people living in today’s digital world. Booking a dinner reservation? Buying something online? (As it happens, Amazon has blocked Muse purchases on its site for the moment, saying that the access violates its terms of service; Meta says Muse can make purchases on other online sites such as Walmart and Expedia.)

Everyone will have to make their own decision about whether the services Muse provides are worth the risks that Broderick cited. Nor should anyone be fooled by the smiley-faced icon Meta chose to represent Muse. For all anyone knows, it’s a wolf in furry clothing.

If Silicon Valley has perfected any business model, it’s the one in which it cooks up a new technology and then persuades people they can’t live without it. The newest example is Meta Platforms’ personal artificial intelligence agent, Muse.

Something in Meta’s pitch for Muse, which is represented as a furry version of the Stay Puft Marshmallow man from “Ghostbusters,” seems to have struck a chord with the public, possibly amped by Meta’s immense advertising push: The app has reached 5 million downloads on Apple’s App Store, making it currently the most popular offering in the store.

Meta makes enticing claims for Muse’s capabilities. The company boasts that Muse can read and write answers to your emails. One of its TV commercials has it sorting the user’s emails to prioritize “what’s important,” placing a carpool schedule on the family calendar, ordering school supplies and turning on her stereo for a house party. The user is shown devoting her free time to chopping vegetables for said party.

Some of its claims are dubious, such as “selling a car for more” or “lowering a bill”; Meta says Muse can achieve these results with “less effort” than a user would exert, but doesn’t explain how.

And some are disquieting. “Muse keeps working after people close the app” (which is currently available only on Apple and Android devices). Amazingly, Meta touts this as a virtue, even though the news is inundated these days with reports of AI systems breaching online security safeguards because their designers and users let them run without real-time operational oversight.

An AI bot that goes on its merry way and informs the user what it’s been up to only when it chooses to? Not the key to a happy relationship.

Meta isn’t alone in trying to convince ordinary people that AI is indispensable. AI companies are encouraging people to upload their health information to be scrutinized by their bots, or to grant access to their bank or investment accounts. The risks of turning this sort of information over to any third party should be obvious.

Meta touts Muse as a free service, but that’s misleading. The company says it expects to make a profit by charging fees for Muse transactions — but as consumers have learned from the proliferation of retailer fees for credit card purchases, those charges can end up costing the consumers themselves. Meta also says Muse is free only up to a certain level of usage, after which users will have to buy a subscription for up to $100 a month. I asked Meta how much usage would trigger the subscription requirement, but it didn’t give me an answer.

Two major issues have already surfaced among Muse’s early users. One is why anyone living a normal life needs a bot to perform such quotidian tasks as reading and writing emails, placing appointments on a calendar or calling a plumber.

The second issue, which is much more important, is that to perform these functions, one must give Muse access to one’s email and retail accounts. Let’s start with that. The opinion of privacy advocates who have weighed in on this requirement generally boils down to: Anyone doing this should have their head examined.

It’s not merely that security breaches exposing the personal information of up to hundreds of millions of people have become almost daily occurrences. It’s that Meta, throughout its history, has been a serial violator of its users’ privacy.

Things came to a head in 2019, when the Federal Trade Commission imposed a record $5-billion fine on the company to settle charges that it had been “deceiving users about their ability to control the privacy of their personal information.” What was striking about that case was that the FTC brought it because it caught the company, then known as Facebook, violating a 2012 consent order that involved similar accusations. As big as the fine appeared, at the time, Facebook collected as much in revenue every month; last year, it collected that much every nine days.

Meta’s misuse of its users’ information was addressed in the $17-billion payment it agreed to in August to settle accusations from California and other states that it knew that features of Facebook and Instagram exposed children using those services to bullies, extortionists and sexual predators.

In the words of California Atty. Gen. Rob Bonta, the company “designed and deployed harmful features on Instagram and Facebook that drive compulsive use by children and teens to their mental and physical detriment, all the while misleading users, their families, and the public regarding the existence and severity of these risks.” (The $17-billion settlement is equivalent to Meta’s revenue for a single month in 2025.)

The company didn’t formally admit to the charges in these cases, but one would have to be hopelessly naive to think that means it’s innocent.

Put all that together and “there’s nothing about Meta’s track record that builds anything near trust that they are going to be responsible stewards or custodians of personal information,” said Lee Hepner, senior legal counsel at the American Economic Liberties Project. “I can’t think of a company with a deeper record of betraying the public trust.”

The defense given by Mark Zuckerberg, the company’s chief executive, chairman and co-founder, of his company’s historical nonchalance about privacy violations has been a sort of shrugging acknowledgment that this is what happens when your focus is on making users’ world “a little better.” In the past, he has copped to his wrongdoing and promised not to do it again, pleading that “we really messed this one up,” “we simply did a bad job with this release,” or “sometimes we move too fast.”

Meta wants us to think that it has learned its lesson. In response to my question about its record on privacy, a company spokesperson referred me to a company post asserting that its Muse agent is endowed with “first-of-its-kind privacy, safety, and security protections … that no other agent provides,” and that “Muse has no visibility into people’s passwords or payment methods. … Muse can use them without seeing them.”

Unexpected leakages of private information already have been cited by early Muse users. Tech commentator Matt Robb reported that Muse gave a would-be buyer of an item he posted for sale a price below his specification and gave the buyer his home address. (The Muse team attributed the glitch to confusing settings in Robb’s Muse account.)

Jason Aten, a tech columnist at Inc. magazine, reported that his Muse read and regurgitated his private messages even though he believed he had denied it that capability. Muse engineers said the bot had exploited a permissions setting on his Mac computer he didn’t realize was applicable.

Those cases point to complexities in the Muse settings that confound experts, let alone laypersons. That undermines Meta’s claim that Muse is “simple to use. People just tell Muse what needs to get done, and it takes action.”

That leaves us with the question of why anyone needs Muse. To answer that question for myself, I launched the sign-up process via my Facebook account, but since the first thing it asked (after prompting me to give the little creature a name) was access to my Gmail account, I shut it down. But other fairly savvy users among tech mavens don’t point to urgent needs that only Muse can satisfy.

Allison Johnson of the Verge writes that her bot successfully emailed a junkman to remove a fallen tree branch in her yard, helped her order camping equipment and put a part for her espresso machine in her online cart but couldn’t complete the purchase. New York Times writer Eli Tan used it to call his dental insurer and wait on hold for a human representative to come on. It also arranged his credit card statements in a spreadsheet and helped him pursue his hobby of collecting ceramic bowls, a task which strikes me as having a fairly small community of potential users.

Muse informed Ryan Broderick, the proprietor of tech blog Garbage Day, that “it could shop for me, book dinner reservations, or … even do tasks on my laptop. But I didn’t need to do any of those things yesterday. And I definitely didn’t need to do any of them so badly that I was willing to risk it deleting something important, hallucinating an email to someone, or buying me something I didn’t want.”

That’s an understandable take. None of the chores that Meta boasts about are beyond the capabilities of most people living in today’s digital world. Booking a dinner reservation? Buying something online? (As it happens, Amazon has blocked Muse purchases on its site for the moment, saying that the access violates its terms of service; Meta says Muse can make purchases on other online sites such as Walmart and Expedia.)

Everyone will have to make their own decision about whether the services Muse provides are worth the risks that Broderick cited. Nor should anyone be fooled by the smiley-faced icon Meta chose to represent Muse. For all anyone knows, it’s a wolf in furry clothing.

If Silicon Valley has perfected any business model, it’s the one in which it cooks up a new technology and then persuades people they can’t live without it. The newest example is Meta Platforms’ personal artificial intelligence agent, Muse.

Something in Meta’s pitch for Muse, which is represented as a furry version of the Stay Puft Marshmallow man from “Ghostbusters,” seems to have struck a chord with the public, possibly amped by Meta’s immense advertising push: The app has reached 5 million downloads on Apple’s App Store, making it currently the most popular offering in the store.

Meta makes enticing claims for Muse’s capabilities. The company boasts that Muse can read and write answers to your emails. One of its TV commercials has it sorting the user’s emails to prioritize “what’s important,” placing a carpool schedule on the family calendar, ordering school supplies and turning on her stereo for a house party. The user is shown devoting her free time to chopping vegetables for said party.

Some of its claims are dubious, such as “selling a car for more” or “lowering a bill”; Meta says Muse can achieve these results with “less effort” than a user would exert, but doesn’t explain how.

And some are disquieting. “Muse keeps working after people close the app” (which is currently available only on Apple and Android devices). Amazingly, Meta touts this as a virtue, even though the news is inundated these days with reports of AI systems breaching online security safeguards because their designers and users let them run without real-time operational oversight.

An AI bot that goes on its merry way and informs the user what it’s been up to only when it chooses to? Not the key to a happy relationship.

Meta isn’t alone in trying to convince ordinary people that AI is indispensable. AI companies are encouraging people to upload their health information to be scrutinized by their bots, or to grant access to their bank or investment accounts. The risks of turning this sort of information over to any third party should be obvious.

Meta touts Muse as a free service, but that’s misleading. The company says it expects to make a profit by charging fees for Muse transactions — but as consumers have learned from the proliferation of retailer fees for credit card purchases, those charges can end up costing the consumers themselves. Meta also says Muse is free only up to a certain level of usage, after which users will have to buy a subscription for up to $100 a month. I asked Meta how much usage would trigger the subscription requirement, but it didn’t give me an answer.

Two major issues have already surfaced among Muse’s early users. One is why anyone living a normal life needs a bot to perform such quotidian tasks as reading and writing emails, placing appointments on a calendar or calling a plumber.

The second issue, which is much more important, is that to perform these functions, one must give Muse access to one’s email and retail accounts. Let’s start with that. The opinion of privacy advocates who have weighed in on this requirement generally boils down to: Anyone doing this should have their head examined.

It’s not merely that security breaches exposing the personal information of up to hundreds of millions of people have become almost daily occurrences. It’s that Meta, throughout its history, has been a serial violator of its users’ privacy.

Things came to a head in 2019, when the Federal Trade Commission imposed a record $5-billion fine on the company to settle charges that it had been “deceiving users about their ability to control the privacy of their personal information.” What was striking about that case was that the FTC brought it because it caught the company, then known as Facebook, violating a 2012 consent order that involved similar accusations. As big as the fine appeared, at the time, Facebook collected as much in revenue every month; last year, it collected that much every nine days.

Meta’s misuse of its users’ information was addressed in the $17-billion payment it agreed to in August to settle accusations from California and other states that it knew that features of Facebook and Instagram exposed children using those services to bullies, extortionists and sexual predators.

In the words of California Atty. Gen. Rob Bonta, the company “designed and deployed harmful features on Instagram and Facebook that drive compulsive use by children and teens to their mental and physical detriment, all the while misleading users, their families, and the public regarding the existence and severity of these risks.” (The $17-billion settlement is equivalent to Meta’s revenue for a single month in 2025.)

The company didn’t formally admit to the charges in these cases, but one would have to be hopelessly naive to think that means it’s innocent.

Put all that together and “there’s nothing about Meta’s track record that builds anything near trust that they are going to be responsible stewards or custodians of personal information,” said Lee Hepner, senior legal counsel at the American Economic Liberties Project. “I can’t think of a company with a deeper record of betraying the public trust.”

The defense given by Mark Zuckerberg, the company’s chief executive, chairman and co-founder, of his company’s historical nonchalance about privacy violations has been a sort of shrugging acknowledgment that this is what happens when your focus is on making users’ world “a little better.” In the past, he has copped to his wrongdoing and promised not to do it again, pleading that “we really messed this one up,” “we simply did a bad job with this release,” or “sometimes we move too fast.”

Meta wants us to think that it has learned its lesson. In response to my question about its record on privacy, a company spokesperson referred me to a company post asserting that its Muse agent is endowed with “first-of-its-kind privacy, safety, and security protections … that no other agent provides,” and that “Muse has no visibility into people’s passwords or payment methods. … Muse can use them without seeing them.”

Unexpected leakages of private information already have been cited by early Muse users. Tech commentator Matt Robb reported that Muse gave a would-be buyer of an item he posted for sale a price below his specification and gave the buyer his home address. (The Muse team attributed the glitch to confusing settings in Robb’s Muse account.)

Jason Aten, a tech columnist at Inc. magazine, reported that his Muse read and regurgitated his private messages even though he believed he had denied it that capability. Muse engineers said the bot had exploited a permissions setting on his Mac computer he didn’t realize was applicable.

Those cases point to complexities in the Muse settings that confound experts, let alone laypersons. That undermines Meta’s claim that Muse is “simple to use. People just tell Muse what needs to get done, and it takes action.”

That leaves us with the question of why anyone needs Muse. To answer that question for myself, I launched the sign-up process via my Facebook account, but since the first thing it asked (after prompting me to give the little creature a name) was access to my Gmail account, I shut it down. But other fairly savvy users among tech mavens don’t point to urgent needs that only Muse can satisfy.

Allison Johnson of the Verge writes that her bot successfully emailed a junkman to remove a fallen tree branch in her yard, helped her order camping equipment and put a part for her espresso machine in her online cart but couldn’t complete the purchase. New York Times writer Eli Tan used it to call his dental insurer and wait on hold for a human representative to come on. It also arranged his credit card statements in a spreadsheet and helped him pursue his hobby of collecting ceramic bowls, a task which strikes me as having a fairly small community of potential users.

Muse informed Ryan Broderick, the proprietor of tech blog Garbage Day, that “it could shop for me, book dinner reservations, or … even do tasks on my laptop. But I didn’t need to do any of those things yesterday. And I definitely didn’t need to do any of them so badly that I was willing to risk it deleting something important, hallucinating an email to someone, or buying me something I didn’t want.”

That’s an understandable take. None of the chores that Meta boasts about are beyond the capabilities of most people living in today’s digital world. Booking a dinner reservation? Buying something online? (As it happens, Amazon has blocked Muse purchases on its site for the moment, saying that the access violates its terms of service; Meta says Muse can make purchases on other online sites such as Walmart and Expedia.)

Everyone will have to make their own decision about whether the services Muse provides are worth the risks that Broderick cited. Nor should anyone be fooled by the smiley-faced icon Meta chose to represent Muse. For all anyone knows, it’s a wolf in furry clothing.

If Silicon Valley has perfected any business model, it’s the one in which it cooks up a new technology and then persuades people they can’t live without it. The newest example is Meta Platforms’ personal artificial intelligence agent, Muse.

Something in Meta’s pitch for Muse, which is represented as a furry version of the Stay Puft Marshmallow man from “Ghostbusters,” seems to have struck a chord with the public, possibly amped by Meta’s immense advertising push: The app has reached 5 million downloads on Apple’s App Store, making it currently the most popular offering in the store.

Meta makes enticing claims for Muse’s capabilities. The company boasts that Muse can read and write answers to your emails. One of its TV commercials has it sorting the user’s emails to prioritize “what’s important,” placing a carpool schedule on the family calendar, ordering school supplies and turning on her stereo for a house party. The user is shown devoting her free time to chopping vegetables for said party.

Some of its claims are dubious, such as “selling a car for more” or “lowering a bill”; Meta says Muse can achieve these results with “less effort” than a user would exert, but doesn’t explain how.

And some are disquieting. “Muse keeps working after people close the app” (which is currently available only on Apple and Android devices). Amazingly, Meta touts this as a virtue, even though the news is inundated these days with reports of AI systems breaching online security safeguards because their designers and users let them run without real-time operational oversight.

An AI bot that goes on its merry way and informs the user what it’s been up to only when it chooses to? Not the key to a happy relationship.

Meta isn’t alone in trying to convince ordinary people that AI is indispensable. AI companies are encouraging people to upload their health information to be scrutinized by their bots, or to grant access to their bank or investment accounts. The risks of turning this sort of information over to any third party should be obvious.

Meta touts Muse as a free service, but that’s misleading. The company says it expects to make a profit by charging fees for Muse transactions — but as consumers have learned from the proliferation of retailer fees for credit card purchases, those charges can end up costing the consumers themselves. Meta also says Muse is free only up to a certain level of usage, after which users will have to buy a subscription for up to $100 a month. I asked Meta how much usage would trigger the subscription requirement, but it didn’t give me an answer.

Two major issues have already surfaced among Muse’s early users. One is why anyone living a normal life needs a bot to perform such quotidian tasks as reading and writing emails, placing appointments on a calendar or calling a plumber.

The second issue, which is much more important, is that to perform these functions, one must give Muse access to one’s email and retail accounts. Let’s start with that. The opinion of privacy advocates who have weighed in on this requirement generally boils down to: Anyone doing this should have their head examined.

It’s not merely that security breaches exposing the personal information of up to hundreds of millions of people have become almost daily occurrences. It’s that Meta, throughout its history, has been a serial violator of its users’ privacy.

Things came to a head in 2019, when the Federal Trade Commission imposed a record $5-billion fine on the company to settle charges that it had been “deceiving users about their ability to control the privacy of their personal information.” What was striking about that case was that the FTC brought it because it caught the company, then known as Facebook, violating a 2012 consent order that involved similar accusations. As big as the fine appeared, at the time, Facebook collected as much in revenue every month; last year, it collected that much every nine days.

Meta’s misuse of its users’ information was addressed in the $17-billion payment it agreed to in August to settle accusations from California and other states that it knew that features of Facebook and Instagram exposed children using those services to bullies, extortionists and sexual predators.

In the words of California Atty. Gen. Rob Bonta, the company “designed and deployed harmful features on Instagram and Facebook that drive compulsive use by children and teens to their mental and physical detriment, all the while misleading users, their families, and the public regarding the existence and severity of these risks.” (The $17-billion settlement is equivalent to Meta’s revenue for a single month in 2025.)

The company didn’t formally admit to the charges in these cases, but one would have to be hopelessly naive to think that means it’s innocent.

Put all that together and “there’s nothing about Meta’s track record that builds anything near trust that they are going to be responsible stewards or custodians of personal information,” said Lee Hepner, senior legal counsel at the American Economic Liberties Project. “I can’t think of a company with a deeper record of betraying the public trust.”

The defense given by Mark Zuckerberg, the company’s chief executive, chairman and co-founder, of his company’s historical nonchalance about privacy violations has been a sort of shrugging acknowledgment that this is what happens when your focus is on making users’ world “a little better.” In the past, he has copped to his wrongdoing and promised not to do it again, pleading that “we really messed this one up,” “we simply did a bad job with this release,” or “sometimes we move too fast.”

Meta wants us to think that it has learned its lesson. In response to my question about its record on privacy, a company spokesperson referred me to a company post asserting that its Muse agent is endowed with “first-of-its-kind privacy, safety, and security protections … that no other agent provides,” and that “Muse has no visibility into people’s passwords or payment methods. … Muse can use them without seeing them.”

Unexpected leakages of private information already have been cited by early Muse users. Tech commentator Matt Robb reported that Muse gave a would-be buyer of an item he posted for sale a price below his specification and gave the buyer his home address. (The Muse team attributed the glitch to confusing settings in Robb’s Muse account.)

Jason Aten, a tech columnist at Inc. magazine, reported that his Muse read and regurgitated his private messages even though he believed he had denied it that capability. Muse engineers said the bot had exploited a permissions setting on his Mac computer he didn’t realize was applicable.

Those cases point to complexities in the Muse settings that confound experts, let alone laypersons. That undermines Meta’s claim that Muse is “simple to use. People just tell Muse what needs to get done, and it takes action.”

That leaves us with the question of why anyone needs Muse. To answer that question for myself, I launched the sign-up process via my Facebook account, but since the first thing it asked (after prompting me to give the little creature a name) was access to my Gmail account, I shut it down. But other fairly savvy users among tech mavens don’t point to urgent needs that only Muse can satisfy.

Allison Johnson of the Verge writes that her bot successfully emailed a junkman to remove a fallen tree branch in her yard, helped her order camping equipment and put a part for her espresso machine in her online cart but couldn’t complete the purchase. New York Times writer Eli Tan used it to call his dental insurer and wait on hold for a human representative to come on. It also arranged his credit card statements in a spreadsheet and helped him pursue his hobby of collecting ceramic bowls, a task which strikes me as having a fairly small community of potential users.

Muse informed Ryan Broderick, the proprietor of tech blog Garbage Day, that “it could shop for me, book dinner reservations, or … even do tasks on my laptop. But I didn’t need to do any of those things yesterday. And I definitely didn’t need to do any of them so badly that I was willing to risk it deleting something important, hallucinating an email to someone, or buying me something I didn’t want.”

That’s an understandable take. None of the chores that Meta boasts about are beyond the capabilities of most people living in today’s digital world. Booking a dinner reservation? Buying something online? (As it happens, Amazon has blocked Muse purchases on its site for the moment, saying that the access violates its terms of service; Meta says Muse can make purchases on other online sites such as Walmart and Expedia.)

Everyone will have to make their own decision about whether the services Muse provides are worth the risks that Broderick cited. Nor should anyone be fooled by the smiley-faced icon Meta chose to represent Muse. For all anyone knows, it’s a wolf in furry clothing.

If Silicon Valley has perfected any business model, it’s the one in which it cooks up a new technology and then persuades people they can’t live without it. The newest example is Meta Platforms’ personal artificial intelligence agent, Muse.

Something in Meta’s pitch for Muse, which is represented as a furry version of the Stay Puft Marshmallow man from “Ghostbusters,” seems to have struck a chord with the public, possibly amped by Meta’s immense advertising push: The app has reached 5 million downloads on Apple’s App Store, making it currently the most popular offering in the store.

Meta makes enticing claims for Muse’s capabilities. The company boasts that Muse can read and write answers to your emails. One of its TV commercials has it sorting the user’s emails to prioritize “what’s important,” placing a carpool schedule on the family calendar, ordering school supplies and turning on her stereo for a house party. The user is shown devoting her free time to chopping vegetables for said party.

Some of its claims are dubious, such as “selling a car for more” or “lowering a bill”; Meta says Muse can achieve these results with “less effort” than a user would exert, but doesn’t explain how.

And some are disquieting. “Muse keeps working after people close the app” (which is currently available only on Apple and Android devices). Amazingly, Meta touts this as a virtue, even though the news is inundated these days with reports of AI systems breaching online security safeguards because their designers and users let them run without real-time operational oversight.

An AI bot that goes on its merry way and informs the user what it’s been up to only when it chooses to? Not the key to a happy relationship.

Meta isn’t alone in trying to convince ordinary people that AI is indispensable. AI companies are encouraging people to upload their health information to be scrutinized by their bots, or to grant access to their bank or investment accounts. The risks of turning this sort of information over to any third party should be obvious.

Meta touts Muse as a free service, but that’s misleading. The company says it expects to make a profit by charging fees for Muse transactions — but as consumers have learned from the proliferation of retailer fees for credit card purchases, those charges can end up costing the consumers themselves. Meta also says Muse is free only up to a certain level of usage, after which users will have to buy a subscription for up to $100 a month. I asked Meta how much usage would trigger the subscription requirement, but it didn’t give me an answer.

Two major issues have already surfaced among Muse’s early users. One is why anyone living a normal life needs a bot to perform such quotidian tasks as reading and writing emails, placing appointments on a calendar or calling a plumber.

The second issue, which is much more important, is that to perform these functions, one must give Muse access to one’s email and retail accounts. Let’s start with that. The opinion of privacy advocates who have weighed in on this requirement generally boils down to: Anyone doing this should have their head examined.

It’s not merely that security breaches exposing the personal information of up to hundreds of millions of people have become almost daily occurrences. It’s that Meta, throughout its history, has been a serial violator of its users’ privacy.

Things came to a head in 2019, when the Federal Trade Commission imposed a record $5-billion fine on the company to settle charges that it had been “deceiving users about their ability to control the privacy of their personal information.” What was striking about that case was that the FTC brought it because it caught the company, then known as Facebook, violating a 2012 consent order that involved similar accusations. As big as the fine appeared, at the time, Facebook collected as much in revenue every month; last year, it collected that much every nine days.

Meta’s misuse of its users’ information was addressed in the $17-billion payment it agreed to in August to settle accusations from California and other states that it knew that features of Facebook and Instagram exposed children using those services to bullies, extortionists and sexual predators.

In the words of California Atty. Gen. Rob Bonta, the company “designed and deployed harmful features on Instagram and Facebook that drive compulsive use by children and teens to their mental and physical detriment, all the while misleading users, their families, and the public regarding the existence and severity of these risks.” (The $17-billion settlement is equivalent to Meta’s revenue for a single month in 2025.)

The company didn’t formally admit to the charges in these cases, but one would have to be hopelessly naive to think that means it’s innocent.

Put all that together and “there’s nothing about Meta’s track record that builds anything near trust that they are going to be responsible stewards or custodians of personal information,” said Lee Hepner, senior legal counsel at the American Economic Liberties Project. “I can’t think of a company with a deeper record of betraying the public trust.”

The defense given by Mark Zuckerberg, the company’s chief executive, chairman and co-founder, of his company’s historical nonchalance about privacy violations has been a sort of shrugging acknowledgment that this is what happens when your focus is on making users’ world “a little better.” In the past, he has copped to his wrongdoing and promised not to do it again, pleading that “we really messed this one up,” “we simply did a bad job with this release,” or “sometimes we move too fast.”

Meta wants us to think that it has learned its lesson. In response to my question about its record on privacy, a company spokesperson referred me to a company post asserting that its Muse agent is endowed with “first-of-its-kind privacy, safety, and security protections … that no other agent provides,” and that “Muse has no visibility into people’s passwords or payment methods. … Muse can use them without seeing them.”

Unexpected leakages of private information already have been cited by early Muse users. Tech commentator Matt Robb reported that Muse gave a would-be buyer of an item he posted for sale a price below his specification and gave the buyer his home address. (The Muse team attributed the glitch to confusing settings in Robb’s Muse account.)

Jason Aten, a tech columnist at Inc. magazine, reported that his Muse read and regurgitated his private messages even though he believed he had denied it that capability. Muse engineers said the bot had exploited a permissions setting on his Mac computer he didn’t realize was applicable.

Those cases point to complexities in the Muse settings that confound experts, let alone laypersons. That undermines Meta’s claim that Muse is “simple to use. People just tell Muse what needs to get done, and it takes action.”

That leaves us with the question of why anyone needs Muse. To answer that question for myself, I launched the sign-up process via my Facebook account, but since the first thing it asked (after prompting me to give the little creature a name) was access to my Gmail account, I shut it down. But other fairly savvy users among tech mavens don’t point to urgent needs that only Muse can satisfy.

Allison Johnson of the Verge writes that her bot successfully emailed a junkman to remove a fallen tree branch in her yard, helped her order camping equipment and put a part for her espresso machine in her online cart but couldn’t complete the purchase. New York Times writer Eli Tan used it to call his dental insurer and wait on hold for a human representative to come on. It also arranged his credit card statements in a spreadsheet and helped him pursue his hobby of collecting ceramic bowls, a task which strikes me as having a fairly small community of potential users.

Muse informed Ryan Broderick, the proprietor of tech blog Garbage Day, that “it could shop for me, book dinner reservations, or … even do tasks on my laptop. But I didn’t need to do any of those things yesterday. And I definitely didn’t need to do any of them so badly that I was willing to risk it deleting something important, hallucinating an email to someone, or buying me something I didn’t want.”

That’s an understandable take. None of the chores that Meta boasts about are beyond the capabilities of most people living in today’s digital world. Booking a dinner reservation? Buying something online? (As it happens, Amazon has blocked Muse purchases on its site for the moment, saying that the access violates its terms of service; Meta says Muse can make purchases on other online sites such as Walmart and Expedia.)

Everyone will have to make their own decision about whether the services Muse provides are worth the risks that Broderick cited. Nor should anyone be fooled by the smiley-faced icon Meta chose to represent Muse. For all anyone knows, it’s a wolf in furry clothing.

If Silicon Valley has perfected any business model, it’s the one in which it cooks up a new technology and then persuades people they can’t live without it. The newest example is Meta Platforms’ personal artificial intelligence agent, Muse.

Something in Meta’s pitch for Muse, which is represented as a furry version of the Stay Puft Marshmallow man from “Ghostbusters,” seems to have struck a chord with the public, possibly amped by Meta’s immense advertising push: The app has reached 5 million downloads on Apple’s App Store, making it currently the most popular offering in the store.

Meta makes enticing claims for Muse’s capabilities. The company boasts that Muse can read and write answers to your emails. One of its TV commercials has it sorting the user’s emails to prioritize “what’s important,” placing a carpool schedule on the family calendar, ordering school supplies and turning on her stereo for a house party. The user is shown devoting her free time to chopping vegetables for said party.

Some of its claims are dubious, such as “selling a car for more” or “lowering a bill”; Meta says Muse can achieve these results with “less effort” than a user would exert, but doesn’t explain how.

And some are disquieting. “Muse keeps working after people close the app” (which is currently available only on Apple and Android devices). Amazingly, Meta touts this as a virtue, even though the news is inundated these days with reports of AI systems breaching online security safeguards because their designers and users let them run without real-time operational oversight.

An AI bot that goes on its merry way and informs the user what it’s been up to only when it chooses to? Not the key to a happy relationship.

Meta isn’t alone in trying to convince ordinary people that AI is indispensable. AI companies are encouraging people to upload their health information to be scrutinized by their bots, or to grant access to their bank or investment accounts. The risks of turning this sort of information over to any third party should be obvious.

Meta touts Muse as a free service, but that’s misleading. The company says it expects to make a profit by charging fees for Muse transactions — but as consumers have learned from the proliferation of retailer fees for credit card purchases, those charges can end up costing the consumers themselves. Meta also says Muse is free only up to a certain level of usage, after which users will have to buy a subscription for up to $100 a month. I asked Meta how much usage would trigger the subscription requirement, but it didn’t give me an answer.

Two major issues have already surfaced among Muse’s early users. One is why anyone living a normal life needs a bot to perform such quotidian tasks as reading and writing emails, placing appointments on a calendar or calling a plumber.

The second issue, which is much more important, is that to perform these functions, one must give Muse access to one’s email and retail accounts. Let’s start with that. The opinion of privacy advocates who have weighed in on this requirement generally boils down to: Anyone doing this should have their head examined.

It’s not merely that security breaches exposing the personal information of up to hundreds of millions of people have become almost daily occurrences. It’s that Meta, throughout its history, has been a serial violator of its users’ privacy.

Things came to a head in 2019, when the Federal Trade Commission imposed a record $5-billion fine on the company to settle charges that it had been “deceiving users about their ability to control the privacy of their personal information.” What was striking about that case was that the FTC brought it because it caught the company, then known as Facebook, violating a 2012 consent order that involved similar accusations. As big as the fine appeared, at the time, Facebook collected as much in revenue every month; last year, it collected that much every nine days.

Meta’s misuse of its users’ information was addressed in the $17-billion payment it agreed to in August to settle accusations from California and other states that it knew that features of Facebook and Instagram exposed children using those services to bullies, extortionists and sexual predators.

In the words of California Atty. Gen. Rob Bonta, the company “designed and deployed harmful features on Instagram and Facebook that drive compulsive use by children and teens to their mental and physical detriment, all the while misleading users, their families, and the public regarding the existence and severity of these risks.” (The $17-billion settlement is equivalent to Meta’s revenue for a single month in 2025.)

The company didn’t formally admit to the charges in these cases, but one would have to be hopelessly naive to think that means it’s innocent.

Put all that together and “there’s nothing about Meta’s track record that builds anything near trust that they are going to be responsible stewards or custodians of personal information,” said Lee Hepner, senior legal counsel at the American Economic Liberties Project. “I can’t think of a company with a deeper record of betraying the public trust.”

The defense given by Mark Zuckerberg, the company’s chief executive, chairman and co-founder, of his company’s historical nonchalance about privacy violations has been a sort of shrugging acknowledgment that this is what happens when your focus is on making users’ world “a little better.” In the past, he has copped to his wrongdoing and promised not to do it again, pleading that “we really messed this one up,” “we simply did a bad job with this release,” or “sometimes we move too fast.”

Meta wants us to think that it has learned its lesson. In response to my question about its record on privacy, a company spokesperson referred me to a company post asserting that its Muse agent is endowed with “first-of-its-kind privacy, safety, and security protections … that no other agent provides,” and that “Muse has no visibility into people’s passwords or payment methods. … Muse can use them without seeing them.”

Unexpected leakages of private information already have been cited by early Muse users. Tech commentator Matt Robb reported that Muse gave a would-be buyer of an item he posted for sale a price below his specification and gave the buyer his home address. (The Muse team attributed the glitch to confusing settings in Robb’s Muse account.)

Jason Aten, a tech columnist at Inc. magazine, reported that his Muse read and regurgitated his private messages even though he believed he had denied it that capability. Muse engineers said the bot had exploited a permissions setting on his Mac computer he didn’t realize was applicable.

Those cases point to complexities in the Muse settings that confound experts, let alone laypersons. That undermines Meta’s claim that Muse is “simple to use. People just tell Muse what needs to get done, and it takes action.”

That leaves us with the question of why anyone needs Muse. To answer that question for myself, I launched the sign-up process via my Facebook account, but since the first thing it asked (after prompting me to give the little creature a name) was access to my Gmail account, I shut it down. But other fairly savvy users among tech mavens don’t point to urgent needs that only Muse can satisfy.

Allison Johnson of the Verge writes that her bot successfully emailed a junkman to remove a fallen tree branch in her yard, helped her order camping equipment and put a part for her espresso machine in her online cart but couldn’t complete the purchase. New York Times writer Eli Tan used it to call his dental insurer and wait on hold for a human representative to come on. It also arranged his credit card statements in a spreadsheet and helped him pursue his hobby of collecting ceramic bowls, a task which strikes me as having a fairly small community of potential users.

Muse informed Ryan Broderick, the proprietor of tech blog Garbage Day, that “it could shop for me, book dinner reservations, or … even do tasks on my laptop. But I didn’t need to do any of those things yesterday. And I definitely didn’t need to do any of them so badly that I was willing to risk it deleting something important, hallucinating an email to someone, or buying me something I didn’t want.”

That’s an understandable take. None of the chores that Meta boasts about are beyond the capabilities of most people living in today’s digital world. Booking a dinner reservation? Buying something online? (As it happens, Amazon has blocked Muse purchases on its site for the moment, saying that the access violates its terms of service; Meta says Muse can make purchases on other online sites such as Walmart and Expedia.)

Everyone will have to make their own decision about whether the services Muse provides are worth the risks that Broderick cited. Nor should anyone be fooled by the smiley-faced icon Meta chose to represent Muse. For all anyone knows, it’s a wolf in furry clothing.

If Silicon Valley has perfected any business model, it’s the one in which it cooks up a new technology and then persuades people they can’t live without it. The newest example is Meta Platforms’ personal artificial intelligence agent, Muse.

Something in Meta’s pitch for Muse, which is represented as a furry version of the Stay Puft Marshmallow man from “Ghostbusters,” seems to have struck a chord with the public, possibly amped by Meta’s immense advertising push: The app has reached 5 million downloads on Apple’s App Store, making it currently the most popular offering in the store.

Meta makes enticing claims for Muse’s capabilities. The company boasts that Muse can read and write answers to your emails. One of its TV commercials has it sorting the user’s emails to prioritize “what’s important,” placing a carpool schedule on the family calendar, ordering school supplies and turning on her stereo for a house party. The user is shown devoting her free time to chopping vegetables for said party.

Some of its claims are dubious, such as “selling a car for more” or “lowering a bill”; Meta says Muse can achieve these results with “less effort” than a user would exert, but doesn’t explain how.

And some are disquieting. “Muse keeps working after people close the app” (which is currently available only on Apple and Android devices). Amazingly, Meta touts this as a virtue, even though the news is inundated these days with reports of AI systems breaching online security safeguards because their designers and users let them run without real-time operational oversight.

An AI bot that goes on its merry way and informs the user what it’s been up to only when it chooses to? Not the key to a happy relationship.

Meta isn’t alone in trying to convince ordinary people that AI is indispensable. AI companies are encouraging people to upload their health information to be scrutinized by their bots, or to grant access to their bank or investment accounts. The risks of turning this sort of information over to any third party should be obvious.

Meta touts Muse as a free service, but that’s misleading. The company says it expects to make a profit by charging fees for Muse transactions — but as consumers have learned from the proliferation of retailer fees for credit card purchases, those charges can end up costing the consumers themselves. Meta also says Muse is free only up to a certain level of usage, after which users will have to buy a subscription for up to $100 a month. I asked Meta how much usage would trigger the subscription requirement, but it didn’t give me an answer.

Two major issues have already surfaced among Muse’s early users. One is why anyone living a normal life needs a bot to perform such quotidian tasks as reading and writing emails, placing appointments on a calendar or calling a plumber.

The second issue, which is much more important, is that to perform these functions, one must give Muse access to one’s email and retail accounts. Let’s start with that. The opinion of privacy advocates who have weighed in on this requirement generally boils down to: Anyone doing this should have their head examined.

It’s not merely that security breaches exposing the personal information of up to hundreds of millions of people have become almost daily occurrences. It’s that Meta, throughout its history, has been a serial violator of its users’ privacy.

Things came to a head in 2019, when the Federal Trade Commission imposed a record $5-billion fine on the company to settle charges that it had been “deceiving users about their ability to control the privacy of their personal information.” What was striking about that case was that the FTC brought it because it caught the company, then known as Facebook, violating a 2012 consent order that involved similar accusations. As big as the fine appeared, at the time, Facebook collected as much in revenue every month; last year, it collected that much every nine days.

Meta’s misuse of its users’ information was addressed in the $17-billion payment it agreed to in August to settle accusations from California and other states that it knew that features of Facebook and Instagram exposed children using those services to bullies, extortionists and sexual predators.

In the words of California Atty. Gen. Rob Bonta, the company “designed and deployed harmful features on Instagram and Facebook that drive compulsive use by children and teens to their mental and physical detriment, all the while misleading users, their families, and the public regarding the existence and severity of these risks.” (The $17-billion settlement is equivalent to Meta’s revenue for a single month in 2025.)

The company didn’t formally admit to the charges in these cases, but one would have to be hopelessly naive to think that means it’s innocent.

Put all that together and “there’s nothing about Meta’s track record that builds anything near trust that they are going to be responsible stewards or custodians of personal information,” said Lee Hepner, senior legal counsel at the American Economic Liberties Project. “I can’t think of a company with a deeper record of betraying the public trust.”

The defense given by Mark Zuckerberg, the company’s chief executive, chairman and co-founder, of his company’s historical nonchalance about privacy violations has been a sort of shrugging acknowledgment that this is what happens when your focus is on making users’ world “a little better.” In the past, he has copped to his wrongdoing and promised not to do it again, pleading that “we really messed this one up,” “we simply did a bad job with this release,” or “sometimes we move too fast.”

Meta wants us to think that it has learned its lesson. In response to my question about its record on privacy, a company spokesperson referred me to a company post asserting that its Muse agent is endowed with “first-of-its-kind privacy, safety, and security protections … that no other agent provides,” and that “Muse has no visibility into people’s passwords or payment methods. … Muse can use them without seeing them.”

Unexpected leakages of private information already have been cited by early Muse users. Tech commentator Matt Robb reported that Muse gave a would-be buyer of an item he posted for sale a price below his specification and gave the buyer his home address. (The Muse team attributed the glitch to confusing settings in Robb’s Muse account.)

Jason Aten, a tech columnist at Inc. magazine, reported that his Muse read and regurgitated his private messages even though he believed he had denied it that capability. Muse engineers said the bot had exploited a permissions setting on his Mac computer he didn’t realize was applicable.

Those cases point to complexities in the Muse settings that confound experts, let alone laypersons. That undermines Meta’s claim that Muse is “simple to use. People just tell Muse what needs to get done, and it takes action.”

That leaves us with the question of why anyone needs Muse. To answer that question for myself, I launched the sign-up process via my Facebook account, but since the first thing it asked (after prompting me to give the little creature a name) was access to my Gmail account, I shut it down. But other fairly savvy users among tech mavens don’t point to urgent needs that only Muse can satisfy.

Allison Johnson of the Verge writes that her bot successfully emailed a junkman to remove a fallen tree branch in her yard, helped her order camping equipment and put a part for her espresso machine in her online cart but couldn’t complete the purchase. New York Times writer Eli Tan used it to call his dental insurer and wait on hold for a human representative to come on. It also arranged his credit card statements in a spreadsheet and helped him pursue his hobby of collecting ceramic bowls, a task which strikes me as having a fairly small community of potential users.

Muse informed Ryan Broderick, the proprietor of tech blog Garbage Day, that “it could shop for me, book dinner reservations, or … even do tasks on my laptop. But I didn’t need to do any of those things yesterday. And I definitely didn’t need to do any of them so badly that I was willing to risk it deleting something important, hallucinating an email to someone, or buying me something I didn’t want.”

That’s an understandable take. None of the chores that Meta boasts about are beyond the capabilities of most people living in today’s digital world. Booking a dinner reservation? Buying something online? (As it happens, Amazon has blocked Muse purchases on its site for the moment, saying that the access violates its terms of service; Meta says Muse can make purchases on other online sites such as Walmart and Expedia.)

Everyone will have to make their own decision about whether the services Muse provides are worth the risks that Broderick cited. Nor should anyone be fooled by the smiley-faced icon Meta chose to represent Muse. For all anyone knows, it’s a wolf in furry clothing.

If Silicon Valley has perfected any business model, it’s the one in which it cooks up a new technology and then persuades people they can’t live without it. The newest example is Meta Platforms’ personal artificial intelligence agent, Muse.

Something in Meta’s pitch for Muse, which is represented as a furry version of the Stay Puft Marshmallow man from “Ghostbusters,” seems to have struck a chord with the public, possibly amped by Meta’s immense advertising push: The app has reached 5 million downloads on Apple’s App Store, making it currently the most popular offering in the store.

Meta makes enticing claims for Muse’s capabilities. The company boasts that Muse can read and write answers to your emails. One of its TV commercials has it sorting the user’s emails to prioritize “what’s important,” placing a carpool schedule on the family calendar, ordering school supplies and turning on her stereo for a house party. The user is shown devoting her free time to chopping vegetables for said party.

Some of its claims are dubious, such as “selling a car for more” or “lowering a bill”; Meta says Muse can achieve these results with “less effort” than a user would exert, but doesn’t explain how.

And some are disquieting. “Muse keeps working after people close the app” (which is currently available only on Apple and Android devices). Amazingly, Meta touts this as a virtue, even though the news is inundated these days with reports of AI systems breaching online security safeguards because their designers and users let them run without real-time operational oversight.

An AI bot that goes on its merry way and informs the user what it’s been up to only when it chooses to? Not the key to a happy relationship.

Meta isn’t alone in trying to convince ordinary people that AI is indispensable. AI companies are encouraging people to upload their health information to be scrutinized by their bots, or to grant access to their bank or investment accounts. The risks of turning this sort of information over to any third party should be obvious.

Meta touts Muse as a free service, but that’s misleading. The company says it expects to make a profit by charging fees for Muse transactions — but as consumers have learned from the proliferation of retailer fees for credit card purchases, those charges can end up costing the consumers themselves. Meta also says Muse is free only up to a certain level of usage, after which users will have to buy a subscription for up to $100 a month. I asked Meta how much usage would trigger the subscription requirement, but it didn’t give me an answer.

Two major issues have already surfaced among Muse’s early users. One is why anyone living a normal life needs a bot to perform such quotidian tasks as reading and writing emails, placing appointments on a calendar or calling a plumber.

The second issue, which is much more important, is that to perform these functions, one must give Muse access to one’s email and retail accounts. Let’s start with that. The opinion of privacy advocates who have weighed in on this requirement generally boils down to: Anyone doing this should have their head examined.

It’s not merely that security breaches exposing the personal information of up to hundreds of millions of people have become almost daily occurrences. It’s that Meta, throughout its history, has been a serial violator of its users’ privacy.

Things came to a head in 2019, when the Federal Trade Commission imposed a record $5-billion fine on the company to settle charges that it had been “deceiving users about their ability to control the privacy of their personal information.” What was striking about that case was that the FTC brought it because it caught the company, then known as Facebook, violating a 2012 consent order that involved similar accusations. As big as the fine appeared, at the time, Facebook collected as much in revenue every month; last year, it collected that much every nine days.

Meta’s misuse of its users’ information was addressed in the $17-billion payment it agreed to in August to settle accusations from California and other states that it knew that features of Facebook and Instagram exposed children using those services to bullies, extortionists and sexual predators.

In the words of California Atty. Gen. Rob Bonta, the company “designed and deployed harmful features on Instagram and Facebook that drive compulsive use by children and teens to their mental and physical detriment, all the while misleading users, their families, and the public regarding the existence and severity of these risks.” (The $17-billion settlement is equivalent to Meta’s revenue for a single month in 2025.)

The company didn’t formally admit to the charges in these cases, but one would have to be hopelessly naive to think that means it’s innocent.

Put all that together and “there’s nothing about Meta’s track record that builds anything near trust that they are going to be responsible stewards or custodians of personal information,” said Lee Hepner, senior legal counsel at the American Economic Liberties Project. “I can’t think of a company with a deeper record of betraying the public trust.”

The defense given by Mark Zuckerberg, the company’s chief executive, chairman and co-founder, of his company’s historical nonchalance about privacy violations has been a sort of shrugging acknowledgment that this is what happens when your focus is on making users’ world “a little better.” In the past, he has copped to his wrongdoing and promised not to do it again, pleading that “we really messed this one up,” “we simply did a bad job with this release,” or “sometimes we move too fast.”

Meta wants us to think that it has learned its lesson. In response to my question about its record on privacy, a company spokesperson referred me to a company post asserting that its Muse agent is endowed with “first-of-its-kind privacy, safety, and security protections … that no other agent provides,” and that “Muse has no visibility into people’s passwords or payment methods. … Muse can use them without seeing them.”

Unexpected leakages of private information already have been cited by early Muse users. Tech commentator Matt Robb reported that Muse gave a would-be buyer of an item he posted for sale a price below his specification and gave the buyer his home address. (The Muse team attributed the glitch to confusing settings in Robb’s Muse account.)

Jason Aten, a tech columnist at Inc. magazine, reported that his Muse read and regurgitated his private messages even though he believed he had denied it that capability. Muse engineers said the bot had exploited a permissions setting on his Mac computer he didn’t realize was applicable.

Those cases point to complexities in the Muse settings that confound experts, let alone laypersons. That undermines Meta’s claim that Muse is “simple to use. People just tell Muse what needs to get done, and it takes action.”

That leaves us with the question of why anyone needs Muse. To answer that question for myself, I launched the sign-up process via my Facebook account, but since the first thing it asked (after prompting me to give the little creature a name) was access to my Gmail account, I shut it down. But other fairly savvy users among tech mavens don’t point to urgent needs that only Muse can satisfy.

Allison Johnson of the Verge writes that her bot successfully emailed a junkman to remove a fallen tree branch in her yard, helped her order camping equipment and put a part for her espresso machine in her online cart but couldn’t complete the purchase. New York Times writer Eli Tan used it to call his dental insurer and wait on hold for a human representative to come on. It also arranged his credit card statements in a spreadsheet and helped him pursue his hobby of collecting ceramic bowls, a task which strikes me as having a fairly small community of potential users.

Muse informed Ryan Broderick, the proprietor of tech blog Garbage Day, that “it could shop for me, book dinner reservations, or … even do tasks on my laptop. But I didn’t need to do any of those things yesterday. And I definitely didn’t need to do any of them so badly that I was willing to risk it deleting something important, hallucinating an email to someone, or buying me something I didn’t want.”

That’s an understandable take. None of the chores that Meta boasts about are beyond the capabilities of most people living in today’s digital world. Booking a dinner reservation? Buying something online? (As it happens, Amazon has blocked Muse purchases on its site for the moment, saying that the access violates its terms of service; Meta says Muse can make purchases on other online sites such as Walmart and Expedia.)

Everyone will have to make their own decision about whether the services Muse provides are worth the risks that Broderick cited. Nor should anyone be fooled by the smiley-faced icon Meta chose to represent Muse. For all anyone knows, it’s a wolf in furry clothing.

If Silicon Valley has perfected any business model, it’s the one in which it cooks up a new technology and then persuades people they can’t live without it. The newest example is Meta Platforms’ personal artificial intelligence agent, Muse.

Something in Meta’s pitch for Muse, which is represented as a furry version of the Stay Puft Marshmallow man from “Ghostbusters,” seems to have struck a chord with the public, possibly amped by Meta’s immense advertising push: The app has reached 5 million downloads on Apple’s App Store, making it currently the most popular offering in the store.

Meta makes enticing claims for Muse’s capabilities. The company boasts that Muse can read and write answers to your emails. One of its TV commercials has it sorting the user’s emails to prioritize “what’s important,” placing a carpool schedule on the family calendar, ordering school supplies and turning on her stereo for a house party. The user is shown devoting her free time to chopping vegetables for said party.

Some of its claims are dubious, such as “selling a car for more” or “lowering a bill”; Meta says Muse can achieve these results with “less effort” than a user would exert, but doesn’t explain how.

And some are disquieting. “Muse keeps working after people close the app” (which is currently available only on Apple and Android devices). Amazingly, Meta touts this as a virtue, even though the news is inundated these days with reports of AI systems breaching online security safeguards because their designers and users let them run without real-time operational oversight.

An AI bot that goes on its merry way and informs the user what it’s been up to only when it chooses to? Not the key to a happy relationship.

Meta isn’t alone in trying to convince ordinary people that AI is indispensable. AI companies are encouraging people to upload their health information to be scrutinized by their bots, or to grant access to their bank or investment accounts. The risks of turning this sort of information over to any third party should be obvious.

Meta touts Muse as a free service, but that’s misleading. The company says it expects to make a profit by charging fees for Muse transactions — but as consumers have learned from the proliferation of retailer fees for credit card purchases, those charges can end up costing the consumers themselves. Meta also says Muse is free only up to a certain level of usage, after which users will have to buy a subscription for up to $100 a month. I asked Meta how much usage would trigger the subscription requirement, but it didn’t give me an answer.

Two major issues have already surfaced among Muse’s early users. One is why anyone living a normal life needs a bot to perform such quotidian tasks as reading and writing emails, placing appointments on a calendar or calling a plumber.

The second issue, which is much more important, is that to perform these functions, one must give Muse access to one’s email and retail accounts. Let’s start with that. The opinion of privacy advocates who have weighed in on this requirement generally boils down to: Anyone doing this should have their head examined.

It’s not merely that security breaches exposing the personal information of up to hundreds of millions of people have become almost daily occurrences. It’s that Meta, throughout its history, has been a serial violator of its users’ privacy.

Things came to a head in 2019, when the Federal Trade Commission imposed a record $5-billion fine on the company to settle charges that it had been “deceiving users about their ability to control the privacy of their personal information.” What was striking about that case was that the FTC brought it because it caught the company, then known as Facebook, violating a 2012 consent order that involved similar accusations. As big as the fine appeared, at the time, Facebook collected as much in revenue every month; last year, it collected that much every nine days.

Meta’s misuse of its users’ information was addressed in the $17-billion payment it agreed to in August to settle accusations from California and other states that it knew that features of Facebook and Instagram exposed children using those services to bullies, extortionists and sexual predators.

In the words of California Atty. Gen. Rob Bonta, the company “designed and deployed harmful features on Instagram and Facebook that drive compulsive use by children and teens to their mental and physical detriment, all the while misleading users, their families, and the public regarding the existence and severity of these risks.” (The $17-billion settlement is equivalent to Meta’s revenue for a single month in 2025.)

The company didn’t formally admit to the charges in these cases, but one would have to be hopelessly naive to think that means it’s innocent.

Put all that together and “there’s nothing about Meta’s track record that builds anything near trust that they are going to be responsible stewards or custodians of personal information,” said Lee Hepner, senior legal counsel at the American Economic Liberties Project. “I can’t think of a company with a deeper record of betraying the public trust.”

The defense given by Mark Zuckerberg, the company’s chief executive, chairman and co-founder, of his company’s historical nonchalance about privacy violations has been a sort of shrugging acknowledgment that this is what happens when your focus is on making users’ world “a little better.” In the past, he has copped to his wrongdoing and promised not to do it again, pleading that “we really messed this one up,” “we simply did a bad job with this release,” or “sometimes we move too fast.”

Meta wants us to think that it has learned its lesson. In response to my question about its record on privacy, a company spokesperson referred me to a company post asserting that its Muse agent is endowed with “first-of-its-kind privacy, safety, and security protections … that no other agent provides,” and that “Muse has no visibility into people’s passwords or payment methods. … Muse can use them without seeing them.”

Unexpected leakages of private information already have been cited by early Muse users. Tech commentator Matt Robb reported that Muse gave a would-be buyer of an item he posted for sale a price below his specification and gave the buyer his home address. (The Muse team attributed the glitch to confusing settings in Robb’s Muse account.)

Jason Aten, a tech columnist at Inc. magazine, reported that his Muse read and regurgitated his private messages even though he believed he had denied it that capability. Muse engineers said the bot had exploited a permissions setting on his Mac computer he didn’t realize was applicable.

Those cases point to complexities in the Muse settings that confound experts, let alone laypersons. That undermines Meta’s claim that Muse is “simple to use. People just tell Muse what needs to get done, and it takes action.”

That leaves us with the question of why anyone needs Muse. To answer that question for myself, I launched the sign-up process via my Facebook account, but since the first thing it asked (after prompting me to give the little creature a name) was access to my Gmail account, I shut it down. But other fairly savvy users among tech mavens don’t point to urgent needs that only Muse can satisfy.

Allison Johnson of the Verge writes that her bot successfully emailed a junkman to remove a fallen tree branch in her yard, helped her order camping equipment and put a part for her espresso machine in her online cart but couldn’t complete the purchase. New York Times writer Eli Tan used it to call his dental insurer and wait on hold for a human representative to come on. It also arranged his credit card statements in a spreadsheet and helped him pursue his hobby of collecting ceramic bowls, a task which strikes me as having a fairly small community of potential users.

Muse informed Ryan Broderick, the proprietor of tech blog Garbage Day, that “it could shop for me, book dinner reservations, or … even do tasks on my laptop. But I didn’t need to do any of those things yesterday. And I definitely didn’t need to do any of them so badly that I was willing to risk it deleting something important, hallucinating an email to someone, or buying me something I didn’t want.”

That’s an understandable take. None of the chores that Meta boasts about are beyond the capabilities of most people living in today’s digital world. Booking a dinner reservation? Buying something online? (As it happens, Amazon has blocked Muse purchases on its site for the moment, saying that the access violates its terms of service; Meta says Muse can make purchases on other online sites such as Walmart and Expedia.)

Everyone will have to make their own decision about whether the services Muse provides are worth the risks that Broderick cited. Nor should anyone be fooled by the smiley-faced icon Meta chose to represent Muse. For all anyone knows, it’s a wolf in furry clothing.

If Silicon Valley has perfected any business model, it’s the one in which it cooks up a new technology and then persuades people they can’t live without it. The newest example is Meta Platforms’ personal artificial intelligence agent, Muse.

Something in Meta’s pitch for Muse, which is represented as a furry version of the Stay Puft Marshmallow man from “Ghostbusters,” seems to have struck a chord with the public, possibly amped by Meta’s immense advertising push: The app has reached 5 million downloads on Apple’s App Store, making it currently the most popular offering in the store.

Meta makes enticing claims for Muse’s capabilities. The company boasts that Muse can read and write answers to your emails. One of its TV commercials has it sorting the user’s emails to prioritize “what’s important,” placing a carpool schedule on the family calendar, ordering school supplies and turning on her stereo for a house party. The user is shown devoting her free time to chopping vegetables for said party.

Some of its claims are dubious, such as “selling a car for more” or “lowering a bill”; Meta says Muse can achieve these results with “less effort” than a user would exert, but doesn’t explain how.

And some are disquieting. “Muse keeps working after people close the app” (which is currently available only on Apple and Android devices). Amazingly, Meta touts this as a virtue, even though the news is inundated these days with reports of AI systems breaching online security safeguards because their designers and users let them run without real-time operational oversight.

An AI bot that goes on its merry way and informs the user what it’s been up to only when it chooses to? Not the key to a happy relationship.

Meta isn’t alone in trying to convince ordinary people that AI is indispensable. AI companies are encouraging people to upload their health information to be scrutinized by their bots, or to grant access to their bank or investment accounts. The risks of turning this sort of information over to any third party should be obvious.

Meta touts Muse as a free service, but that’s misleading. The company says it expects to make a profit by charging fees for Muse transactions — but as consumers have learned from the proliferation of retailer fees for credit card purchases, those charges can end up costing the consumers themselves. Meta also says Muse is free only up to a certain level of usage, after which users will have to buy a subscription for up to $100 a month. I asked Meta how much usage would trigger the subscription requirement, but it didn’t give me an answer.

Two major issues have already surfaced among Muse’s early users. One is why anyone living a normal life needs a bot to perform such quotidian tasks as reading and writing emails, placing appointments on a calendar or calling a plumber.

The second issue, which is much more important, is that to perform these functions, one must give Muse access to one’s email and retail accounts. Let’s start with that. The opinion of privacy advocates who have weighed in on this requirement generally boils down to: Anyone doing this should have their head examined.

It’s not merely that security breaches exposing the personal information of up to hundreds of millions of people have become almost daily occurrences. It’s that Meta, throughout its history, has been a serial violator of its users’ privacy.

Things came to a head in 2019, when the Federal Trade Commission imposed a record $5-billion fine on the company to settle charges that it had been “deceiving users about their ability to control the privacy of their personal information.” What was striking about that case was that the FTC brought it because it caught the company, then known as Facebook, violating a 2012 consent order that involved similar accusations. As big as the fine appeared, at the time, Facebook collected as much in revenue every month; last year, it collected that much every nine days.

Meta’s misuse of its users’ information was addressed in the $17-billion payment it agreed to in August to settle accusations from California and other states that it knew that features of Facebook and Instagram exposed children using those services to bullies, extortionists and sexual predators.

In the words of California Atty. Gen. Rob Bonta, the company “designed and deployed harmful features on Instagram and Facebook that drive compulsive use by children and teens to their mental and physical detriment, all the while misleading users, their families, and the public regarding the existence and severity of these risks.” (The $17-billion settlement is equivalent to Meta’s revenue for a single month in 2025.)

The company didn’t formally admit to the charges in these cases, but one would have to be hopelessly naive to think that means it’s innocent.

Put all that together and “there’s nothing about Meta’s track record that builds anything near trust that they are going to be responsible stewards or custodians of personal information,” said Lee Hepner, senior legal counsel at the American Economic Liberties Project. “I can’t think of a company with a deeper record of betraying the public trust.”

The defense given by Mark Zuckerberg, the company’s chief executive, chairman and co-founder, of his company’s historical nonchalance about privacy violations has been a sort of shrugging acknowledgment that this is what happens when your focus is on making users’ world “a little better.” In the past, he has copped to his wrongdoing and promised not to do it again, pleading that “we really messed this one up,” “we simply did a bad job with this release,” or “sometimes we move too fast.”

Meta wants us to think that it has learned its lesson. In response to my question about its record on privacy, a company spokesperson referred me to a company post asserting that its Muse agent is endowed with “first-of-its-kind privacy, safety, and security protections … that no other agent provides,” and that “Muse has no visibility into people’s passwords or payment methods. … Muse can use them without seeing them.”

Unexpected leakages of private information already have been cited by early Muse users. Tech commentator Matt Robb reported that Muse gave a would-be buyer of an item he posted for sale a price below his specification and gave the buyer his home address. (The Muse team attributed the glitch to confusing settings in Robb’s Muse account.)

Jason Aten, a tech columnist at Inc. magazine, reported that his Muse read and regurgitated his private messages even though he believed he had denied it that capability. Muse engineers said the bot had exploited a permissions setting on his Mac computer he didn’t realize was applicable.

Those cases point to complexities in the Muse settings that confound experts, let alone laypersons. That undermines Meta’s claim that Muse is “simple to use. People just tell Muse what needs to get done, and it takes action.”

That leaves us with the question of why anyone needs Muse. To answer that question for myself, I launched the sign-up process via my Facebook account, but since the first thing it asked (after prompting me to give the little creature a name) was access to my Gmail account, I shut it down. But other fairly savvy users among tech mavens don’t point to urgent needs that only Muse can satisfy.

Allison Johnson of the Verge writes that her bot successfully emailed a junkman to remove a fallen tree branch in her yard, helped her order camping equipment and put a part for her espresso machine in her online cart but couldn’t complete the purchase. New York Times writer Eli Tan used it to call his dental insurer and wait on hold for a human representative to come on. It also arranged his credit card statements in a spreadsheet and helped him pursue his hobby of collecting ceramic bowls, a task which strikes me as having a fairly small community of potential users.

Muse informed Ryan Broderick, the proprietor of tech blog Garbage Day, that “it could shop for me, book dinner reservations, or … even do tasks on my laptop. But I didn’t need to do any of those things yesterday. And I definitely didn’t need to do any of them so badly that I was willing to risk it deleting something important, hallucinating an email to someone, or buying me something I didn’t want.”

That’s an understandable take. None of the chores that Meta boasts about are beyond the capabilities of most people living in today’s digital world. Booking a dinner reservation? Buying something online? (As it happens, Amazon has blocked Muse purchases on its site for the moment, saying that the access violates its terms of service; Meta says Muse can make purchases on other online sites such as Walmart and Expedia.)

Everyone will have to make their own decision about whether the services Muse provides are worth the risks that Broderick cited. Nor should anyone be fooled by the smiley-faced icon Meta chose to represent Muse. For all anyone knows, it’s a wolf in furry clothing.

If Silicon Valley has perfected any business model, it’s the one in which it cooks up a new technology and then persuades people they can’t live without it. The newest example is Meta Platforms’ personal artificial intelligence agent, Muse.

Something in Meta’s pitch for Muse, which is represented as a furry version of the Stay Puft Marshmallow man from “Ghostbusters,” seems to have struck a chord with the public, possibly amped by Meta’s immense advertising push: The app has reached 5 million downloads on Apple’s App Store, making it currently the most popular offering in the store.

Meta makes enticing claims for Muse’s capabilities. The company boasts that Muse can read and write answers to your emails. One of its TV commercials has it sorting the user’s emails to prioritize “what’s important,” placing a carpool schedule on the family calendar, ordering school supplies and turning on her stereo for a house party. The user is shown devoting her free time to chopping vegetables for said party.

Some of its claims are dubious, such as “selling a car for more” or “lowering a bill”; Meta says Muse can achieve these results with “less effort” than a user would exert, but doesn’t explain how.

And some are disquieting. “Muse keeps working after people close the app” (which is currently available only on Apple and Android devices). Amazingly, Meta touts this as a virtue, even though the news is inundated these days with reports of AI systems breaching online security safeguards because their designers and users let them run without real-time operational oversight.

An AI bot that goes on its merry way and informs the user what it’s been up to only when it chooses to? Not the key to a happy relationship.

Meta isn’t alone in trying to convince ordinary people that AI is indispensable. AI companies are encouraging people to upload their health information to be scrutinized by their bots, or to grant access to their bank or investment accounts. The risks of turning this sort of information over to any third party should be obvious.

Meta touts Muse as a free service, but that’s misleading. The company says it expects to make a profit by charging fees for Muse transactions — but as consumers have learned from the proliferation of retailer fees for credit card purchases, those charges can end up costing the consumers themselves. Meta also says Muse is free only up to a certain level of usage, after which users will have to buy a subscription for up to $100 a month. I asked Meta how much usage would trigger the subscription requirement, but it didn’t give me an answer.

Two major issues have already surfaced among Muse’s early users. One is why anyone living a normal life needs a bot to perform such quotidian tasks as reading and writing emails, placing appointments on a calendar or calling a plumber.

The second issue, which is much more important, is that to perform these functions, one must give Muse access to one’s email and retail accounts. Let’s start with that. The opinion of privacy advocates who have weighed in on this requirement generally boils down to: Anyone doing this should have their head examined.

It’s not merely that security breaches exposing the personal information of up to hundreds of millions of people have become almost daily occurrences. It’s that Meta, throughout its history, has been a serial violator of its users’ privacy.

Things came to a head in 2019, when the Federal Trade Commission imposed a record $5-billion fine on the company to settle charges that it had been “deceiving users about their ability to control the privacy of their personal information.” What was striking about that case was that the FTC brought it because it caught the company, then known as Facebook, violating a 2012 consent order that involved similar accusations. As big as the fine appeared, at the time, Facebook collected as much in revenue every month; last year, it collected that much every nine days.

Meta’s misuse of its users’ information was addressed in the $17-billion payment it agreed to in August to settle accusations from California and other states that it knew that features of Facebook and Instagram exposed children using those services to bullies, extortionists and sexual predators.

In the words of California Atty. Gen. Rob Bonta, the company “designed and deployed harmful features on Instagram and Facebook that drive compulsive use by children and teens to their mental and physical detriment, all the while misleading users, their families, and the public regarding the existence and severity of these risks.” (The $17-billion settlement is equivalent to Meta’s revenue for a single month in 2025.)

The company didn’t formally admit to the charges in these cases, but one would have to be hopelessly naive to think that means it’s innocent.

Put all that together and “there’s nothing about Meta’s track record that builds anything near trust that they are going to be responsible stewards or custodians of personal information,” said Lee Hepner, senior legal counsel at the American Economic Liberties Project. “I can’t think of a company with a deeper record of betraying the public trust.”

The defense given by Mark Zuckerberg, the company’s chief executive, chairman and co-founder, of his company’s historical nonchalance about privacy violations has been a sort of shrugging acknowledgment that this is what happens when your focus is on making users’ world “a little better.” In the past, he has copped to his wrongdoing and promised not to do it again, pleading that “we really messed this one up,” “we simply did a bad job with this release,” or “sometimes we move too fast.”

Meta wants us to think that it has learned its lesson. In response to my question about its record on privacy, a company spokesperson referred me to a company post asserting that its Muse agent is endowed with “first-of-its-kind privacy, safety, and security protections … that no other agent provides,” and that “Muse has no visibility into people’s passwords or payment methods. … Muse can use them without seeing them.”

Unexpected leakages of private information already have been cited by early Muse users. Tech commentator Matt Robb reported that Muse gave a would-be buyer of an item he posted for sale a price below his specification and gave the buyer his home address. (The Muse team attributed the glitch to confusing settings in Robb’s Muse account.)

Jason Aten, a tech columnist at Inc. magazine, reported that his Muse read and regurgitated his private messages even though he believed he had denied it that capability. Muse engineers said the bot had exploited a permissions setting on his Mac computer he didn’t realize was applicable.

Those cases point to complexities in the Muse settings that confound experts, let alone laypersons. That undermines Meta’s claim that Muse is “simple to use. People just tell Muse what needs to get done, and it takes action.”

That leaves us with the question of why anyone needs Muse. To answer that question for myself, I launched the sign-up process via my Facebook account, but since the first thing it asked (after prompting me to give the little creature a name) was access to my Gmail account, I shut it down. But other fairly savvy users among tech mavens don’t point to urgent needs that only Muse can satisfy.

Allison Johnson of the Verge writes that her bot successfully emailed a junkman to remove a fallen tree branch in her yard, helped her order camping equipment and put a part for her espresso machine in her online cart but couldn’t complete the purchase. New York Times writer Eli Tan used it to call his dental insurer and wait on hold for a human representative to come on. It also arranged his credit card statements in a spreadsheet and helped him pursue his hobby of collecting ceramic bowls, a task which strikes me as having a fairly small community of potential users.

Muse informed Ryan Broderick, the proprietor of tech blog Garbage Day, that “it could shop for me, book dinner reservations, or … even do tasks on my laptop. But I didn’t need to do any of those things yesterday. And I definitely didn’t need to do any of them so badly that I was willing to risk it deleting something important, hallucinating an email to someone, or buying me something I didn’t want.”

That’s an understandable take. None of the chores that Meta boasts about are beyond the capabilities of most people living in today’s digital world. Booking a dinner reservation? Buying something online? (As it happens, Amazon has blocked Muse purchases on its site for the moment, saying that the access violates its terms of service; Meta says Muse can make purchases on other online sites such as Walmart and Expedia.)

Everyone will have to make their own decision about whether the services Muse provides are worth the risks that Broderick cited. Nor should anyone be fooled by the smiley-faced icon Meta chose to represent Muse. For all anyone knows, it’s a wolf in furry clothing.

If Silicon Valley has perfected any business model, it’s the one in which it cooks up a new technology and then persuades people they can’t live without it. The newest example is Meta Platforms’ personal artificial intelligence agent, Muse.

Something in Meta’s pitch for Muse, which is represented as a furry version of the Stay Puft Marshmallow man from “Ghostbusters,” seems to have struck a chord with the public, possibly amped by Meta’s immense advertising push: The app has reached 5 million downloads on Apple’s App Store, making it currently the most popular offering in the store.

Meta makes enticing claims for Muse’s capabilities. The company boasts that Muse can read and write answers to your emails. One of its TV commercials has it sorting the user’s emails to prioritize “what’s important,” placing a carpool schedule on the family calendar, ordering school supplies and turning on her stereo for a house party. The user is shown devoting her free time to chopping vegetables for said party.

Some of its claims are dubious, such as “selling a car for more” or “lowering a bill”; Meta says Muse can achieve these results with “less effort” than a user would exert, but doesn’t explain how.

And some are disquieting. “Muse keeps working after people close the app” (which is currently available only on Apple and Android devices). Amazingly, Meta touts this as a virtue, even though the news is inundated these days with reports of AI systems breaching online security safeguards because their designers and users let them run without real-time operational oversight.

An AI bot that goes on its merry way and informs the user what it’s been up to only when it chooses to? Not the key to a happy relationship.

Meta isn’t alone in trying to convince ordinary people that AI is indispensable. AI companies are encouraging people to upload their health information to be scrutinized by their bots, or to grant access to their bank or investment accounts. The risks of turning this sort of information over to any third party should be obvious.

Meta touts Muse as a free service, but that’s misleading. The company says it expects to make a profit by charging fees for Muse transactions — but as consumers have learned from the proliferation of retailer fees for credit card purchases, those charges can end up costing the consumers themselves. Meta also says Muse is free only up to a certain level of usage, after which users will have to buy a subscription for up to $100 a month. I asked Meta how much usage would trigger the subscription requirement, but it didn’t give me an answer.

Two major issues have already surfaced among Muse’s early users. One is why anyone living a normal life needs a bot to perform such quotidian tasks as reading and writing emails, placing appointments on a calendar or calling a plumber.

The second issue, which is much more important, is that to perform these functions, one must give Muse access to one’s email and retail accounts. Let’s start with that. The opinion of privacy advocates who have weighed in on this requirement generally boils down to: Anyone doing this should have their head examined.

It’s not merely that security breaches exposing the personal information of up to hundreds of millions of people have become almost daily occurrences. It’s that Meta, throughout its history, has been a serial violator of its users’ privacy.

Things came to a head in 2019, when the Federal Trade Commission imposed a record $5-billion fine on the company to settle charges that it had been “deceiving users about their ability to control the privacy of their personal information.” What was striking about that case was that the FTC brought it because it caught the company, then known as Facebook, violating a 2012 consent order that involved similar accusations. As big as the fine appeared, at the time, Facebook collected as much in revenue every month; last year, it collected that much every nine days.

Meta’s misuse of its users’ information was addressed in the $17-billion payment it agreed to in August to settle accusations from California and other states that it knew that features of Facebook and Instagram exposed children using those services to bullies, extortionists and sexual predators.

In the words of California Atty. Gen. Rob Bonta, the company “designed and deployed harmful features on Instagram and Facebook that drive compulsive use by children and teens to their mental and physical detriment, all the while misleading users, their families, and the public regarding the existence and severity of these risks.” (The $17-billion settlement is equivalent to Meta’s revenue for a single month in 2025.)

The company didn’t formally admit to the charges in these cases, but one would have to be hopelessly naive to think that means it’s innocent.

Put all that together and “there’s nothing about Meta’s track record that builds anything near trust that they are going to be responsible stewards or custodians of personal information,” said Lee Hepner, senior legal counsel at the American Economic Liberties Project. “I can’t think of a company with a deeper record of betraying the public trust.”

The defense given by Mark Zuckerberg, the company’s chief executive, chairman and co-founder, of his company’s historical nonchalance about privacy violations has been a sort of shrugging acknowledgment that this is what happens when your focus is on making users’ world “a little better.” In the past, he has copped to his wrongdoing and promised not to do it again, pleading that “we really messed this one up,” “we simply did a bad job with this release,” or “sometimes we move too fast.”

Meta wants us to think that it has learned its lesson. In response to my question about its record on privacy, a company spokesperson referred me to a company post asserting that its Muse agent is endowed with “first-of-its-kind privacy, safety, and security protections … that no other agent provides,” and that “Muse has no visibility into people’s passwords or payment methods. … Muse can use them without seeing them.”

Unexpected leakages of private information already have been cited by early Muse users. Tech commentator Matt Robb reported that Muse gave a would-be buyer of an item he posted for sale a price below his specification and gave the buyer his home address. (The Muse team attributed the glitch to confusing settings in Robb’s Muse account.)

Jason Aten, a tech columnist at Inc. magazine, reported that his Muse read and regurgitated his private messages even though he believed he had denied it that capability. Muse engineers said the bot had exploited a permissions setting on his Mac computer he didn’t realize was applicable.

Those cases point to complexities in the Muse settings that confound experts, let alone laypersons. That undermines Meta’s claim that Muse is “simple to use. People just tell Muse what needs to get done, and it takes action.”

That leaves us with the question of why anyone needs Muse. To answer that question for myself, I launched the sign-up process via my Facebook account, but since the first thing it asked (after prompting me to give the little creature a name) was access to my Gmail account, I shut it down. But other fairly savvy users among tech mavens don’t point to urgent needs that only Muse can satisfy.

Allison Johnson of the Verge writes that her bot successfully emailed a junkman to remove a fallen tree branch in her yard, helped her order camping equipment and put a part for her espresso machine in her online cart but couldn’t complete the purchase. New York Times writer Eli Tan used it to call his dental insurer and wait on hold for a human representative to come on. It also arranged his credit card statements in a spreadsheet and helped him pursue his hobby of collecting ceramic bowls, a task which strikes me as having a fairly small community of potential users.

Muse informed Ryan Broderick, the proprietor of tech blog Garbage Day, that “it could shop for me, book dinner reservations, or … even do tasks on my laptop. But I didn’t need to do any of those things yesterday. And I definitely didn’t need to do any of them so badly that I was willing to risk it deleting something important, hallucinating an email to someone, or buying me something I didn’t want.”

That’s an understandable take. None of the chores that Meta boasts about are beyond the capabilities of most people living in today’s digital world. Booking a dinner reservation? Buying something online? (As it happens, Amazon has blocked Muse purchases on its site for the moment, saying that the access violates its terms of service; Meta says Muse can make purchases on other online sites such as Walmart and Expedia.)

Everyone will have to make their own decision about whether the services Muse provides are worth the risks that Broderick cited. Nor should anyone be fooled by the smiley-faced icon Meta chose to represent Muse. For all anyone knows, it’s a wolf in furry clothing.

If Silicon Valley has perfected any business model, it’s the one in which it cooks up a new technology and then persuades people they can’t live without it. The newest example is Meta Platforms’ personal artificial intelligence agent, Muse.

Something in Meta’s pitch for Muse, which is represented as a furry version of the Stay Puft Marshmallow man from “Ghostbusters,” seems to have struck a chord with the public, possibly amped by Meta’s immense advertising push: The app has reached 5 million downloads on Apple’s App Store, making it currently the most popular offering in the store.

Meta makes enticing claims for Muse’s capabilities. The company boasts that Muse can read and write answers to your emails. One of its TV commercials has it sorting the user’s emails to prioritize “what’s important,” placing a carpool schedule on the family calendar, ordering school supplies and turning on her stereo for a house party. The user is shown devoting her free time to chopping vegetables for said party.

Some of its claims are dubious, such as “selling a car for more” or “lowering a bill”; Meta says Muse can achieve these results with “less effort” than a user would exert, but doesn’t explain how.

And some are disquieting. “Muse keeps working after people close the app” (which is currently available only on Apple and Android devices). Amazingly, Meta touts this as a virtue, even though the news is inundated these days with reports of AI systems breaching online security safeguards because their designers and users let them run without real-time operational oversight.

An AI bot that goes on its merry way and informs the user what it’s been up to only when it chooses to? Not the key to a happy relationship.

Meta isn’t alone in trying to convince ordinary people that AI is indispensable. AI companies are encouraging people to upload their health information to be scrutinized by their bots, or to grant access to their bank or investment accounts. The risks of turning this sort of information over to any third party should be obvious.

Meta touts Muse as a free service, but that’s misleading. The company says it expects to make a profit by charging fees for Muse transactions — but as consumers have learned from the proliferation of retailer fees for credit card purchases, those charges can end up costing the consumers themselves. Meta also says Muse is free only up to a certain level of usage, after which users will have to buy a subscription for up to $100 a month. I asked Meta how much usage would trigger the subscription requirement, but it didn’t give me an answer.

Two major issues have already surfaced among Muse’s early users. One is why anyone living a normal life needs a bot to perform such quotidian tasks as reading and writing emails, placing appointments on a calendar or calling a plumber.

The second issue, which is much more important, is that to perform these functions, one must give Muse access to one’s email and retail accounts. Let’s start with that. The opinion of privacy advocates who have weighed in on this requirement generally boils down to: Anyone doing this should have their head examined.

It’s not merely that security breaches exposing the personal information of up to hundreds of millions of people have become almost daily occurrences. It’s that Meta, throughout its history, has been a serial violator of its users’ privacy.

Things came to a head in 2019, when the Federal Trade Commission imposed a record $5-billion fine on the company to settle charges that it had been “deceiving users about their ability to control the privacy of their personal information.” What was striking about that case was that the FTC brought it because it caught the company, then known as Facebook, violating a 2012 consent order that involved similar accusations. As big as the fine appeared, at the time, Facebook collected as much in revenue every month; last year, it collected that much every nine days.

Meta’s misuse of its users’ information was addressed in the $17-billion payment it agreed to in August to settle accusations from California and other states that it knew that features of Facebook and Instagram exposed children using those services to bullies, extortionists and sexual predators.

In the words of California Atty. Gen. Rob Bonta, the company “designed and deployed harmful features on Instagram and Facebook that drive compulsive use by children and teens to their mental and physical detriment, all the while misleading users, their families, and the public regarding the existence and severity of these risks.” (The $17-billion settlement is equivalent to Meta’s revenue for a single month in 2025.)

The company didn’t formally admit to the charges in these cases, but one would have to be hopelessly naive to think that means it’s innocent.

Put all that together and “there’s nothing about Meta’s track record that builds anything near trust that they are going to be responsible stewards or custodians of personal information,” said Lee Hepner, senior legal counsel at the American Economic Liberties Project. “I can’t think of a company with a deeper record of betraying the public trust.”

The defense given by Mark Zuckerberg, the company’s chief executive, chairman and co-founder, of his company’s historical nonchalance about privacy violations has been a sort of shrugging acknowledgment that this is what happens when your focus is on making users’ world “a little better.” In the past, he has copped to his wrongdoing and promised not to do it again, pleading that “we really messed this one up,” “we simply did a bad job with this release,” or “sometimes we move too fast.”

Meta wants us to think that it has learned its lesson. In response to my question about its record on privacy, a company spokesperson referred me to a company post asserting that its Muse agent is endowed with “first-of-its-kind privacy, safety, and security protections … that no other agent provides,” and that “Muse has no visibility into people’s passwords or payment methods. … Muse can use them without seeing them.”

Unexpected leakages of private information already have been cited by early Muse users. Tech commentator Matt Robb reported that Muse gave a would-be buyer of an item he posted for sale a price below his specification and gave the buyer his home address. (The Muse team attributed the glitch to confusing settings in Robb’s Muse account.)

Jason Aten, a tech columnist at Inc. magazine, reported that his Muse read and regurgitated his private messages even though he believed he had denied it that capability. Muse engineers said the bot had exploited a permissions setting on his Mac computer he didn’t realize was applicable.

Those cases point to complexities in the Muse settings that confound experts, let alone laypersons. That undermines Meta’s claim that Muse is “simple to use. People just tell Muse what needs to get done, and it takes action.”

That leaves us with the question of why anyone needs Muse. To answer that question for myself, I launched the sign-up process via my Facebook account, but since the first thing it asked (after prompting me to give the little creature a name) was access to my Gmail account, I shut it down. But other fairly savvy users among tech mavens don’t point to urgent needs that only Muse can satisfy.

Allison Johnson of the Verge writes that her bot successfully emailed a junkman to remove a fallen tree branch in her yard, helped her order camping equipment and put a part for her espresso machine in her online cart but couldn’t complete the purchase. New York Times writer Eli Tan used it to call his dental insurer and wait on hold for a human representative to come on. It also arranged his credit card statements in a spreadsheet and helped him pursue his hobby of collecting ceramic bowls, a task which strikes me as having a fairly small community of potential users.

Muse informed Ryan Broderick, the proprietor of tech blog Garbage Day, that “it could shop for me, book dinner reservations, or … even do tasks on my laptop. But I didn’t need to do any of those things yesterday. And I definitely didn’t need to do any of them so badly that I was willing to risk it deleting something important, hallucinating an email to someone, or buying me something I didn’t want.”

That’s an understandable take. None of the chores that Meta boasts about are beyond the capabilities of most people living in today’s digital world. Booking a dinner reservation? Buying something online? (As it happens, Amazon has blocked Muse purchases on its site for the moment, saying that the access violates its terms of service; Meta says Muse can make purchases on other online sites such as Walmart and Expedia.)

Everyone will have to make their own decision about whether the services Muse provides are worth the risks that Broderick cited. Nor should anyone be fooled by the smiley-faced icon Meta chose to represent Muse. For all anyone knows, it’s a wolf in furry clothing.

If Silicon Valley has perfected any business model, it’s the one in which it cooks up a new technology and then persuades people they can’t live without it. The newest example is Meta Platforms’ personal artificial intelligence agent, Muse.

Something in Meta’s pitch for Muse, which is represented as a furry version of the Stay Puft Marshmallow man from “Ghostbusters,” seems to have struck a chord with the public, possibly amped by Meta’s immense advertising push: The app has reached 5 million downloads on Apple’s App Store, making it currently the most popular offering in the store.

Meta makes enticing claims for Muse’s capabilities. The company boasts that Muse can read and write answers to your emails. One of its TV commercials has it sorting the user’s emails to prioritize “what’s important,” placing a carpool schedule on the family calendar, ordering school supplies and turning on her stereo for a house party. The user is shown devoting her free time to chopping vegetables for said party.

Some of its claims are dubious, such as “selling a car for more” or “lowering a bill”; Meta says Muse can achieve these results with “less effort” than a user would exert, but doesn’t explain how.

And some are disquieting. “Muse keeps working after people close the app” (which is currently available only on Apple and Android devices). Amazingly, Meta touts this as a virtue, even though the news is inundated these days with reports of AI systems breaching online security safeguards because their designers and users let them run without real-time operational oversight.

An AI bot that goes on its merry way and informs the user what it’s been up to only when it chooses to? Not the key to a happy relationship.

Meta isn’t alone in trying to convince ordinary people that AI is indispensable. AI companies are encouraging people to upload their health information to be scrutinized by their bots, or to grant access to their bank or investment accounts. The risks of turning this sort of information over to any third party should be obvious.

Meta touts Muse as a free service, but that’s misleading. The company says it expects to make a profit by charging fees for Muse transactions — but as consumers have learned from the proliferation of retailer fees for credit card purchases, those charges can end up costing the consumers themselves. Meta also says Muse is free only up to a certain level of usage, after which users will have to buy a subscription for up to $100 a month. I asked Meta how much usage would trigger the subscription requirement, but it didn’t give me an answer.

Two major issues have already surfaced among Muse’s early users. One is why anyone living a normal life needs a bot to perform such quotidian tasks as reading and writing emails, placing appointments on a calendar or calling a plumber.

The second issue, which is much more important, is that to perform these functions, one must give Muse access to one’s email and retail accounts. Let’s start with that. The opinion of privacy advocates who have weighed in on this requirement generally boils down to: Anyone doing this should have their head examined.

It’s not merely that security breaches exposing the personal information of up to hundreds of millions of people have become almost daily occurrences. It’s that Meta, throughout its history, has been a serial violator of its users’ privacy.

Things came to a head in 2019, when the Federal Trade Commission imposed a record $5-billion fine on the company to settle charges that it had been “deceiving users about their ability to control the privacy of their personal information.” What was striking about that case was that the FTC brought it because it caught the company, then known as Facebook, violating a 2012 consent order that involved similar accusations. As big as the fine appeared, at the time, Facebook collected as much in revenue every month; last year, it collected that much every nine days.

Meta’s misuse of its users’ information was addressed in the $17-billion payment it agreed to in August to settle accusations from California and other states that it knew that features of Facebook and Instagram exposed children using those services to bullies, extortionists and sexual predators.

In the words of California Atty. Gen. Rob Bonta, the company “designed and deployed harmful features on Instagram and Facebook that drive compulsive use by children and teens to their mental and physical detriment, all the while misleading users, their families, and the public regarding the existence and severity of these risks.” (The $17-billion settlement is equivalent to Meta’s revenue for a single month in 2025.)

The company didn’t formally admit to the charges in these cases, but one would have to be hopelessly naive to think that means it’s innocent.

Put all that together and “there’s nothing about Meta’s track record that builds anything near trust that they are going to be responsible stewards or custodians of personal information,” said Lee Hepner, senior legal counsel at the American Economic Liberties Project. “I can’t think of a company with a deeper record of betraying the public trust.”

The defense given by Mark Zuckerberg, the company’s chief executive, chairman and co-founder, of his company’s historical nonchalance about privacy violations has been a sort of shrugging acknowledgment that this is what happens when your focus is on making users’ world “a little better.” In the past, he has copped to his wrongdoing and promised not to do it again, pleading that “we really messed this one up,” “we simply did a bad job with this release,” or “sometimes we move too fast.”

Meta wants us to think that it has learned its lesson. In response to my question about its record on privacy, a company spokesperson referred me to a company post asserting that its Muse agent is endowed with “first-of-its-kind privacy, safety, and security protections … that no other agent provides,” and that “Muse has no visibility into people’s passwords or payment methods. … Muse can use them without seeing them.”

Unexpected leakages of private information already have been cited by early Muse users. Tech commentator Matt Robb reported that Muse gave a would-be buyer of an item he posted for sale a price below his specification and gave the buyer his home address. (The Muse team attributed the glitch to confusing settings in Robb’s Muse account.)

Jason Aten, a tech columnist at Inc. magazine, reported that his Muse read and regurgitated his private messages even though he believed he had denied it that capability. Muse engineers said the bot had exploited a permissions setting on his Mac computer he didn’t realize was applicable.

Those cases point to complexities in the Muse settings that confound experts, let alone laypersons. That undermines Meta’s claim that Muse is “simple to use. People just tell Muse what needs to get done, and it takes action.”

That leaves us with the question of why anyone needs Muse. To answer that question for myself, I launched the sign-up process via my Facebook account, but since the first thing it asked (after prompting me to give the little creature a name) was access to my Gmail account, I shut it down. But other fairly savvy users among tech mavens don’t point to urgent needs that only Muse can satisfy.

Allison Johnson of the Verge writes that her bot successfully emailed a junkman to remove a fallen tree branch in her yard, helped her order camping equipment and put a part for her espresso machine in her online cart but couldn’t complete the purchase. New York Times writer Eli Tan used it to call his dental insurer and wait on hold for a human representative to come on. It also arranged his credit card statements in a spreadsheet and helped him pursue his hobby of collecting ceramic bowls, a task which strikes me as having a fairly small community of potential users.

Muse informed Ryan Broderick, the proprietor of tech blog Garbage Day, that “it could shop for me, book dinner reservations, or … even do tasks on my laptop. But I didn’t need to do any of those things yesterday. And I definitely didn’t need to do any of them so badly that I was willing to risk it deleting something important, hallucinating an email to someone, or buying me something I didn’t want.”

That’s an understandable take. None of the chores that Meta boasts about are beyond the capabilities of most people living in today’s digital world. Booking a dinner reservation? Buying something online? (As it happens, Amazon has blocked Muse purchases on its site for the moment, saying that the access violates its terms of service; Meta says Muse can make purchases on other online sites such as Walmart and Expedia.)

Everyone will have to make their own decision about whether the services Muse provides are worth the risks that Broderick cited. Nor should anyone be fooled by the smiley-faced icon Meta chose to represent Muse. For all anyone knows, it’s a wolf in furry clothing.