Meta employees warned that its Muse AI agent was reading private messages and showing them bank balances that weren’t theirs

by | Nov 19, 2026 | Latest E-commerce News & Updates

Meta employees testing its Muse AI agent warned that it appeared to read private messages users hadn’t shared and to mix up conversations between accounts, according to internal chats reviewed by Reuters. In one case, a staffer chatting about fishing got an unprompted message listing Capital One and Charles Schwab balances that weren’t theirs. Meta said those outputs were isolated hallucinations triggered by built-in prompts and have been fixed, but Reuters couldn’t confirm whether the data was real. When Reuters tested Muse with Signal and the AI Now Institute, the agent read Signal messages on Android using permissions from setup and on macOS through Apple’s accessibility tools. Apple and Google are now weighing changes to give users more control over what AI agents can access.

Paul Drecksler is the founder and editor of Shopifreaks, covering the most important stories in e-commerce.

Companies: Meta

Never miss important e-commerce news

Our weekly newsletter is read religiously by 20,000+ e-commerce professionals.

Loading...