Meta announced it has temporarily disabled the feature that lets users share links to their Muse conversations after a security investigation revealed that the URLs could be accessed without authentication. The decision was reported by Ars Technica and reflects growing scrutiny of AI tools that inadvertently expose user data. The pause applies to all accounts using the Muse assistant.
Claroc, a cybersecurity firm, used public DNS records as a lookup mechanism to enumerate Muse link identifiers. By scanning these records they identified more than 6,800 distinct URLs that pointed to active conversation pages. The firm says the method required only standard DNS queries, demonstrating how routine internet infrastructure can inadvertently reveal private AI interactions when proper safeguards are absent.
Among the publicly reachable chats were documents that included a marketing firm’s quarterly revenue figures, a small-business owner’s loan application, a full-year staff calendar for an organization, and preparatory notes for a candidate interview with an artificial-intelligence laboratory. The presence of such varied and sensitive content illustrates the breadth of information that can be unintentionally exposed when conversation links lack authentication barriers or expiration controls.
Claroc reported that hundreds of these conversations were viewable in plain text, meaning anyone with the URL could read the full dialogue without additional verification. The firm warned that such exposure could facilitate social engineering, corporate espionage, or identity theft, especially when the data includes personal identifiers or proprietary business metrics.
The incident joins earlier cases where AI chat-sharing functions unintentionally indexed content on search engines. Similar leaks have been documented for OpenAI’s ChatGPT, Anthropic’s Claude, and xAI’s Grok, where shared conversation links appeared in Google search results. These precedents highlight a systemic challenge in balancing convenient sharing with robust privacy protections across AI platforms.
In response, Meta indicated the pause will remain until it can implement stronger access controls and possibly introduce link expiration or authentication requirements. Industry observers suggest the episode may prompt broader regulatory attention to AI-generated content sharing, urging developers to embed privacy-by-design principles. The company has not disclosed a timeline for reinstating the feature.