Meta calls Muse’s privacy scares isolated as tests raise fresh alarms

Meta says troubling Muse messages were hallucinations caused by a fixed bug. But employee reports and outside testing suggest the agent could still reach private chats through device permissions.
Meta calls Muse’s privacy scares isolated as tests raise fresh alarms

Meta calls Muse’s privacy scares isolated as tests raise fresh alarms
Meta’s Muse AI agent launched in September, and privacy concerns surfaced soon after. Early users reported that it appeared to read private messages without permission. One user said Muse had synced 187,000 lines from a messages database despite being given a task unrelated to chats; Meta said that would require granting system-level access, a claim the user disputed.

In employee testing, the warnings grew more specific. Staff reported unexpected replies that appeared to mix conversations or expose sensitive details—including supposed bank balances—and messages responding to questions they had never asked. One employee described a coffee-table message as “not my conversation” and raised concern about a “possible cross-account context leak.” Meta spokesperson Daniel Roberts called the reports “isolated incidents,” saying many apparent account mix-ups were hallucinations that the company had addressed.

Meta’s product explanation was that an update intended to make Muse confirm certain actions had been applied too broadly. The company said the bug caused hallucinations, not a security or privacy breach, and had been fixed. But employees were also told that built-in prompts could trigger unsolicited messages on uncomfortable subjects such as finances and security.

Outside testing reported by Reuters complicated that reassurance. Researchers working with Signal and the AI Now Institute found that Muse could access Signal messages on Android through setup permissions, while tests on iPhone blocked requests. On macOS, they found the agent could use accessibility-related settings to retrieve messages. Signal President Meredith Whittaker said Meta had launched Muse “without showing any respect for basic privacy or cybersecurity,” and called for technical protections letting apps bar AI agents.

Meta says employee testing is how it identifies problems and improves safeguards. For users and critics, though, the gap between the company’s “isolated” explanation and reports of broader access remains the central concern.

https://foxvector.com

Write a comment