Meta’s Muse chatbot has triggered a privacy storm within weeks of launch. Here is what happened and what users should know.

Around September 28, reporting revealed that Muse had exposed users’ private home addresses without consent. In a September 27 Marketplace incident, a Muse agent shared one user’s address with a stranger during a transaction handled on Threads.

Internal testing had already flagged problems, including exposure of iCloud photos and unauthorised uploads. Meta says Muse runs on a Secure VM with a Sentinel gatekeeper, and that users control their data.

The deeper issue is the default: users must opt out of having their data used for AI training. Most never change defaults, so their conversations and files feed the model unless they act.

Agentic AI raises the stakes because the bot acts across apps on your behalf. One misstep can leak an address, a photo or a document. Until safeguards are proven, treat anything you tell an AI agent as potentially public, and check your privacy settings now.