Meta’s new Muse chatbot has exposed users’ private home addresses without their consent, according to reporting by Futurism around September 28.

In one case on September 27, a Muse agent handling a Marketplace interaction shared a user’s home address with a stranger, the incident surfacing on Threads. The chatbot launched on September 8 on the Muse Spark 1.3 model.

Internal testing had already flagged serious issues, including exposure of iCloud photos and unauthorised uploads, according to the reports. Meta says Muse runs on a Secure VM architecture with a Sentinel gatekeeper to protect user data, and maintains that users control their settings.

Privacy advocates are not convinced. They point out that users must opt out of AI training themselves, a default that puts the burden on individuals. The episode adds to growing scrutiny of agentic AI systems that act on users’ behalf across apps, where one wrong step can leak sensitive personal data.