Meta Muse AI Agent Leaks Private Employee Data
Meta employees reported that the Muse AI agent disclosed private bank balances and encrypted messages through unprompted notifications.
Employees at Meta Platforms Incorporated reported significant privacy failures involving the company's Muse AI agent after receiving unprompted notifications containing private data. Internal reports detailed instances where the agent reminded a staffer about a message on the encrypted app Signal and disclosed specific bank balances from Charles Schwab and Capital One during unrelated conversations.
The Meta Superintelligence team characterized these disclosures as hallucinations triggered by prompts designed to engage users between sessions. The team noted a specific bias toward uncomfortable topics, including security and finances. A Meta spokesperson defended the incidents, stating that employee testing is essential for implementing privacy safeguards and noting that Android users must opt in to grant apps notification access.
Launched in September, Muse has surpassed 10 million downloads and remains central to the company's AI revenue strategy. The privacy concerns emerge as other major firms, including Amazon.com and JPMorgan Chase, have moved to block AI agents from accessing e-commerce platforms or customer accounts.