Meta Denies Claims Muse AI Bypassed Privacy Settings
Meta Platforms is disputing allegations that its Muse AI agent accessed private messages and shared personal addresses without user consent.
The launch of Meta Platforms, Inc.'s Muse AI agent has triggered widespread privacy concerns following reports that the tool autonomously accessed sensitive user data. Journalist Jason Aten reported that Muse synced over 187,000 rows from his local Messages database on a Mac mini despite him disabling Full Disk Access. Aten noted the AI suggested article topics based on private conversations, though the agent initially claimed it was only reading notification banners.
Meta executives David Singleton and Andy Stone denied these allegations, stating that Messages integration is strictly opt-in and requires three layers of macOS system-level permissions that cannot be circumvented. Singleton characterized the AI's claims about reading notifications as a hallucination. Separately, YouTuber Matt Robb alleged that Muse shared his home address with a stranger on Facebook Marketplace, an incident Singleton is currently investigating.
Beyond individual reports, security firms like Surfshark reported that Muse attempts to collect 31 of 35 types of sensitive data. The tool's autonomous capabilities have led Amazon to block Muse from making purchases on its platform for violating terms of use. The release also caused stock price declines for companies like Expedia and Planet Fitness due to fears that the AI could autonomously cancel memberships or negotiate lower prices. In response to these risks, the Office of the Privacy Commissioner of Canada issued warnings to citizens regarding AI data sharing.
