Meta Denies Its AI Agent Muse Read Users’ Private Messages Without Permission

October 1 (IT Home) — Meta has pushed back against a journalist’s claim that its AI agent Muse read users’ private messages without permission.

IT Home notes that Jason Aten, a columnist for Inc. magazine, previously published a report describing the problem in detail. Afterwards Andy Stone, Meta’s vice president of communications, responded publicly, making clear that the company does not believe its product has read messages without the user’s permission.

In a post on X responding to the report’s claims, Stone said: “The message integration in the Muse app for Mac is entirely opt-in. A user must enable both Full Disk Access and the Messages connector before Muse can read your message content. If you don’t do that, it cannot read your messages.”

Although Meta has denied it, many users still suspect the company is not telling the full story.

Given that this technology giant has repeatedly stumbled over user data handling in the past — leading to lawsuits, Federal Trade Commission findings of violation and a number of fines — such doubt is hardly surprising. Just days ago, a jury in New Mexico ruled in a case that the tech giant had misled users about its data handling; that case grew out of the Cambridge Analytica data scandal of 2018.

Whether users are willing to trust Muse will be a key factor in whether Meta can succeed in the consumer artificial intelligence market. The app is doing well for now and remains at the top of the app store download charts; but if more reports like this appear in future, Meta’s reputation could become difficult to repair, whether or not the reports are accurate. Many observers argue that the company should have spoken directly with the journalist to find out why the situation occurred, rather than simply denying that it happened.

Before Stone issued a formal statement on Meta’s behalf, David Singleton, an executive at Meta Superintelligence Labs, had already replied to Aten directly on Threads. He explained the technical side: for Muse to read a user’s messages on a Mac, the user must complete three separate application permission settings, and macOS itself also has built-in protections. Singleton said that “even if the Muse app itself had a bug, those protections could not be bypassed.”

The steps are as follows: the user must explicitly grant Muse Full Disk Access. Only after that can the user choose Muse’s level of access to the Messages app, with options of no access, read-only, and read. If Full Disk Access is not switched on, those later options appear greyed out and cannot be clicked.

In addition, when Full Disk Access is enabled, a system pop-up brings up the macOS settings interface, where the user must once again manually confirm the action. Singleton wrote that after setup is complete, the Muse app also fully restarts once. This further reduces the chance of a user accidentally granting authorisation without realising it.

Aten’s report claims, however, that Full Disk Access was switched off when Muse read his messages. He also says that when he asked Muse why this had happened, the AI replied that it was syncing the user’s “device notifications.” On that basis, Aten believes Muse submitted the text of banner notifications popping up on the Mac to the AI agent.

Singleton disputes this account as well. He says the AI itself became confused and gave an incorrect explanation of what happened; he then pointed readers to the pages on Meta’s official website describing Muse’s security architecture and its bug bounty programme.

In short, Meta’s position is essentially this: what Aten described neither actually happened nor would be technically possible.

This is not the first controversy over Muse overstepping its authority, and it is unlikely to be the last. The YouTube creator Matt Robb recently reported that while he was selling items on Facebook Marketplace, Muse made a mistake while handling a task that led to his home address being exposed; one buyer even turned up at his door while he was not at home. Meta has opened an investigation into the complaint, and the whole matter is fairly complicated. The user himself also said that he had previously granted Muse a certain permission, which ultimately caused the problem.