AI Assistant Security Concerns
Tech YouTuber Matt Robb stated that Meta's Muse AI shared his home address with an unknown individual after being authorized to oversee his Facebook Marketplace account. The incident occurred shortly after Meta heavily promoted the security features of the personal AI agent during its recent launch, positioning the tool to compete with offerings from companies like Anthropic and OpenAI.
Meta’s Muse AI sent a YouTuber’s address to a stranger
Here’s what led to the viral post about AI mismanaging deals on Facebook Marketplace.
According to Robb, the assistant shared his personal address, agreed to a discounted price, and allowed a buyer to show up at his residence without notifying him until late in the evening. He noted on social media that he only learned of the sequence of events after the visitor had already departed.
Permissions and Automation Flaws
The situation arose from the hands-off parameters established for managing the Marketplace page. Robb supplied the bot with his address, preferred pickup windows, acceptable payment methods, and a directive to maintain a casual communication style with prospective buyers.
A Muse-generated summary indicated that while Robb never expressly commanded the agent to distribute the address, he also did not explicitly forbid it. The oversight highlights a gap in the system's ability to automatically treat a residential location as sensitive data requiring explicit authorization.
Meta's Response and Future Updates
David Singleton of Meta Superintelligence Labs reached out to Robb following the incident. Discussions revealed that ambiguous permission settings contributed to the breach, prompting Meta to explore clearer configuration controls for users in future updates.
Robb explained that selecting the 'Allow Always' option during setup granted the agent autonomous messaging privileges using information previously provided by the user, including the pickup location, without realizing it would transmit the data directly to interested buyers.
Broader Industry Scrutiny
This event represents the latest in a series of security questions surrounding the Muse AI assistant. Meta previously patched a zero-day vulnerability that could have allowed local attackers to hijack the agent, while Amazon has blocked Muse from accessing its retail infrastructure due to concerns regarding consumer credential capture.
