India's news, as it happens

Live TV
Tech

Meta employees report privacy concerns over Muse AI agent

Internal warnings suggest the AI agent may access private messages and provide unsolicited information regarding finances and insurance.

Meta employees testing the company's Muse AI agent have raised concerns that the tool may access encrypted private messages and pull details from other users' conversations, according to a Reuters report cited by the Times of India.

The agent reportedly provided unsolicited information during tests, such as listing specific bank balances at Capital One and Charles Schwab during a conversation about fishing. In another instance, an employee claimed the agent switched their dog to a premium insurance plan costing $888 a month, an increase from $200.

One employee flagged an incident where the agent offered to contact a seller regarding a coffee table pickup that the user had not requested, describing it as a possible cross-account context leak. Additionally, a reporter for Inc Magazine, Jason Aten, claimed the agent accessed private messages on his Mac without permission despite him explicitly denying access to personal information.

Meta spokesperson Daniel Roberts told Reuters that these were isolated incidents. He stated that employees were informed internally that the agent seemed to be generating imaginary messages due to built-in prompts designed to encourage user engagement.

Roberts added that many messages flagged by staff as potential account mix-ups were actually hallucinations that the company had already addressed. He did not respond to questions regarding how common these issues are or if they continue to persist.

Spotted a mistake? Tell us

Related stories

More in Tech

All Tech news