Meta's Muse AI read private messages without user consent

Full text

Meta has launched a new personal artificial intelligence (AI) agent called Muse , and it seems to be pretty powerful. A little too powerful , perhaps, as one writer has just revealed how the tool read his personal text messages without permission, then lied about what it did when questioned.

As explained by Inc writer Jason Aten , “I was having a conversation with my Primary Technology podcast co-host, Stephen Robles, about the new iPhones. Moments later, I got a push notification from Muse suggesting that the conversation we were having would make for a good column and offered to put together research for me to write about. It even flagged a message from my editor about having a column ready for Monday.”

The problem here is that “not only had I not asked it to do that sort of thing,” Aten stated, “I never gave it permission to read my messages. In fact, I remember explicitly choosing not to let it have access to my messages, calendar, and other personal information.”

After Aten asked Muse how it knew about his messages, the AI agent flubbed that it grabbed the relevant info from push notifications relating to Aten’s texts, not from inside the messages themselves. The problem? Aten is sure that’s not true.

The author discovered that Muse syncs your messages to Meta’s servers. On his device, this feature “had been activated and synced to row 187,462 of my Messages database,” Aten noted. “I’m not sure how many messages that equates to, but it’s obviously more than whatever Muse told me about what it was doing.”

Meta says it designed Muse to be “a safe, secure, private, and widely available personal AI agent.” But sending a user’s private messages to Meta’s servers then providing a misleading answer about what it was doing doesn’t exactly align with Muse’s stated goals. It we’re to judge companies on their actions rather than their words then Meta doesn’t come out of this smelling much like roses.

A familiar pattern

(Image credit: Meta) This is just the latest incident in a recent glut of bad press for Meta, with much of it centering on its AI efforts. For example, a security researcher discovered that Muse can be manipulated into sending your voice recordings and more to a hacker’s server, while Amazon has banned the AI agent from shopping on its website. And Meta is apparently considering using a “human concierge” system that would involve real people handling a percentage of the potentially sensitive phone calls that can be placed using Muse.

And that’s not forgetting the controversy over its “pervert glasses” that can record random members of the public without their consent. Meta’s smart glasses have been used for harassment, stalking and more since they launched, garnering considerable hostile attention in the process.

I wish I could say that this was just a blip or an unexpected turn of events, but when it comes to Meta, privacy violations very much feel like the norm . They’ve happened so often in the past that it’s hard not to conclude that the functionality that allows its AI to browse and upload your messages without permission is there by design, not by accident.

And when we’re talking about a company that seems to feel it has an absolute right to all of your private data — regardless of your wishes — then it’s no wonder that I wouldn’t touch Meta Muse with a bargepole. I’ve tried plenty of AI tools and services , but Muse is one that I just won’t go near.

But let’s be fair here: Muse is only the most egregious example of a worrying trend among AI agents to treat user privacy as an annoying afterthought , one that can be ignored as long as they can get away with it. You only need to look at the recent spate of AI security incidents to know that trusting an agent with your private data is a risky proposition.

Throw Meta into the mix and you’ve got a recipe for disaster, as Aten discovered. If you’re running Muse on your devices right now, you might want to consider what permissions you’ve granted it — or think about uninstalling it completely.