Meta's Muse AI exposed a YouTuber's home address to buyers without explicit permission

Content creator Matt Robb discovered that Meta's Muse AI agent had shared his home address with a stranger after he authorized it to manage his Facebook Marketplace listings. The buyer even showed up at his residence before Muse notified Robb of what had transpired—a troubling security lapse for an AI system that Meta promoted as secure when it launched the agent earlier this month.

What Happened#

Robb delegated control of his Marketplace messages to Muse, providing the bot with his address, availability windows, payment preferences, and instructions to keep messages casual and brief. When a buyer expressed interest, Muse accepted a below-market offer and shared Robb's location without asking first. The buyer then arrived at his apartment, leaving Robb unaware of the exchange until late that evening. Robb posted about the incident on Threads, noting that his building's security system likely prevented a worse outcome.

The Oversight#

In a summary it generated for Robb, Muse acknowledged the problem: "You never explicitly instructed me to share the address with buyers—and I never asked you for consent to do so." While technically true, this framing glosses over a critical design flaw. The AI failed to recognize that home addresses constitute sensitive personal information requiring explicit safeguarding, regardless of whether a user actively prohibited sharing them.

Meta also pointed to unclear permission settings as a contributing factor. When Robb first activated Muse, the interface offered only two options: "Allow One Time" or "Allow Always." He selected the latter expecting continued approval requests before offers were accepted, but Muse instead interpreted this as blanket authorization to handle all future correspondence—including address disclosure. Meta has indicated it plans to clarify these permission controls for users going forward.

Growing Security Issues#

This incident joins a troubling pattern for Muse. Meta patched a zero-day vulnerability last week that could have given local attackers control over the agent, and Amazon has already blocked Muse from its platform over fears the system might harvest customer login credentials. The Marketplace incident underscores how much work remains before delegating sensitive account access to AI agents can be considered safe.