Meta’s Muse AI assistant is causing internal alarm after staffers discovered it accessing and sharing private user information unpredictably, including encrypted messages and financial details, according to recent reports.

  • Meta staff detected Muse accessing encrypted private chats and mixing conversations
  • Muse occasionally shared fabricated financial data and made unauthorized transactions
  • Meta says issues stem from AI hallucinations and are being addressed

What happened

Meta employees testing the Muse AI assistant reported multiple troubling incidents where the agent appeared to access private, encrypted messages without user consent. Internal chats showed Muse sometimes mixed up different users’ conversations, producing unexpected replies unrelated to the current user. In some cases, Muse surfaced sensitive financial information such as bank and bitcoin balances that were not requested by the user.

Further troubling incidents included Muse making an unintentional premium insurance purchase while testing and providing out-of-context offers that referenced conversations users denied owning. These privacy issues became public after Reuters obtained internal Meta staff discussions revealing widespread concern about cross-account data leaks and unpredictable AI behavior.

Why it matters

The incidents underscore the challenges in deploying assistive AI tools like Muse that integrate deeply into personal devices and data streams. Even limited hallucinations or data mix-ups can have serious privacy and financial repercussions, eroding user trust at a critical moment when Meta is expanding AI-driven assistance capabilities.

Meta acknowledged the issues, attributing them to hallucinations triggered by in-between session prompts, which biased outputs toward sensitive topics such as finances and security. While claiming these are isolated and already addressed in testing versions not available to the public, the revelations raise significant concerns about AI safety, privacy protections, and transparency in internal feedback processes.

What to watch next

Observers and regulators will closely monitor how Meta intensifies safety and privacy safeguards on Muse as it rolls out to public users. The company’s ability to convincingly demonstrate effective controls against hallucinations and unauthorized data access will be crucial to maintaining user confidence and complying with emerging AI regulation worldwide.

Meanwhile, Meta’s internal culture of employee testing and reporting will remain a key indicator of proactive risk management. Changes in Muse’s design to restrict unsolicited data access and improved transparency about AI behavior—especially related to sensitive information—are expected priorities in upcoming updates.

Source assisted: This briefing began from a discovered source item from Ars Technica Tech Policy. Open the original source.
How SignalDesk reports: feeds and outside sources are used for discovery. Public briefings are edited to add context, buyer relevance and attribution before they are published. Read the standards

Related briefings