Meta’s Muse AI agent allegedly read private messages without permission; Journalist raises privacy concerns

Planning to install AI agents on your personal device? Think twice because it may gather sensitive data, including private messages and other information stored on your device. In a recent controversy, Meta’s Muse AI agent is facing scrutiny over its capabilities and privacy concerns. 

In a recent incident, a technology columnist, Jason Aten, shared a concerning experience with Muse AI, claiming that the agent accessed and synced his private Messages database, including around 187,000 lines of messages, despite Full Disk Access being turned off.

Must read: Meta Muse AI allegedly shared user’s address, finalised a sale deal without approval: Buyer showed up at 9:15 PM

If you’re unaware of Muse AI, then Meta recently released the AI agent that can perform tasks on users’ behalf, such as researching information, creating briefings, and even helping with shopping. It operates via a dedicated virtual computer, with the ability to access information from connected apps or other sources. Meta says Muse follows the permissions set by the user. However, in Aten’s case, the AI agent allegedly worked beyond his permissions. 

Muse red flags around privacy

According to an Inc. report, Aten installed Muse on his iPhone and Mac mini to test how the AI agent works. However, he kept Full Disk Access disabled for privacy reasons, which means the AI can not access his Messages, calendar, or other personal information. 

For testing, he initially asked Muse to write a biography based on what it knew about him, and asked researching story topics, helping with podcast guests and preparing morning briefings. However, Aten caught unusual behaviour when Muse suggested an article based on a conversation he was having with podcast co-host Stephen Robles about the new iPhones.

Must read: OpenAI cancels new AI model launch last-minute over safety concerns: All details

Reportedly, Muse also referenced a message from his editor about having a column ready for Monday. This was unusual for Aten, as he did not allow the AI agent to monitor his messages or suggest stories based on his conversations.

When he asked Muse how it knew, it reportedly told Aten that it wasn't reading his Messages history. Instead, it said the Muse app on his Mac was simply receiving incoming notification banners, which gave snippets of messages. “I only saw the notification preview, not your actual Messages database,” the agent said, according to Aten.

The hidden sync uncovered

Aten was not quite satisfied with Muse’s response, so he decided to investigate and found something quite shocking. Reportedly, Muse had synced data from the local Messages database on his Mac and had reached row 187,462 of that database. Aten clarified that Full Disk Access was not enabled for Muse in his Mac's settings, making it a concerning incident.  While Aten reached out to the Meta team, the company executive David Singleton responded to his Threads post, saying that the feature was opt-in. If users agreed, Facebook could continuously select and upload photos from their camera roll to Meta’s cloud. Meta said it would use information such as the time, location, and themes associated with photos to understand them.

Despite Meta's explanation that the feature requires user consent, the incident highlighted a major flaw in transparency and permission controls. This incident may act as a critical warning for users and regulators regarding AI agents installed on personal devices.