Muse exposes users' personal and business secrets when they share conversations

Meta is suspending a feature in its AI agent Muse that exposed user data, meaning you can no longer share Muse conversations via web links.

Meta's AI agent, Muse, has a critical privacy flaw where shared conversations can expose sensitive personal and business information. Researchers discovered that by querying a specific domain, they could easily enumerate and access thousands of these shared conversations. The exposed data included confidential executive search notes, internal revenue reviews, loan applications, bank fraud alerts, real-estate investment reports, and staff vacation calendars. Meta has acknowledged the issue and has suspended the sharing feature while it investigates and plans potential changes. This vulnerability was reported through Meta's bug bounty program on October 8th. The ease of discovery is attributed to the use of a predictable domain and the inclusion of conversation titles directly in the URL's subdomain.

The ease with which Muse's "GhostShares" were discoverable, primarily through passive DNS records and predictable URL structures including conversation titles in subdomains, highlights a significant oversight in Meta's approach to AI privacy. This incident suggests that despite claims of prioritizing privacy and security, fundamental design decisions in AI product development can inadvertently create large-scale data leakage risks. The company's swift suspension of the feature indicates the severity of the exposure, but the underlying technical oversight raises questions about the security posture of other AI agents that rely on similar sharing mechanisms or require broad access to user data.

This event will likely pressure other AI developers to more rigorously audit their sharing functionalities and consider alternative, more secure methods for content distribution. Users who rely on AI agents for sensitive tasks may need to reassess their trust in these tools and look for explicit guarantees of data protection beyond marketing claims. The incident also underscores the ongoing challenge of balancing user convenience with robust data security in the rapidly evolving AI landscape, potentially leading to industry-wide shifts in how AI-generated content is shared and secured.