Meta Suspends Muse AI Chat Sharing After Public URL Exposure Discovered

Image: Meta pauses Muse sharing after researchers find AI chats on the web · Source

Meta temporarily halted the Muse AI assistant's conversation sharing feature after security researchers found thousands of shared chat links publicly accessible via DNS records, raising privacy concerns.

Meta has suspended a sharing feature of its AI assistant, Muse, following a security firm's discovery that many shared conversation links were publicly accessible online. Claroc, a security research company, reported finding over 6,800 URLs related to Muse conversations by examining public DNS records, which act as web traffic directories. This discovery indicated that shared Muse chat pages could be easily located and accessed by anyone with the URL or by reconstructing the URL from DNS hostnames.

The sharing feature allows Muse users to generate links to their AI conversation threads so others can view them. However, Claroc explained that Meta's design incorporates conversation titles and unique IDs directly into the web addresses, making it straightforward to predict or enumerate URLs based on publicly available DNS information. For example, the hostname within DNS often revealed the nature of the content before the page was even opened, facilitating the retrieval of confidential information.

Among the exposed content were sensitive materials such as a marketing company's revenue reports, a small business owner's loan application details, a full-year staff calendar from a workplace, and notes from a job candidate’s interview with an AI research lab. These findings highlight the potential privacy risks when AI conversation sharing features lack robust access controls.

Tracy Clayton, a spokesperson for Muse, stated in an email to The Verge that Meta has paused this sharing capability to review the issue and consider modifications to prevent easy discovery of shared links. Previous instances have shown similar concerns with AI chat sharing features from other providers like OpenAI’s ChatGPT, Anthropic’s Claude, and xAI’s Grok, where conversations unintentionally appeared in public search results.

Meta's move to temporarily disable Muse’s share links underscores the challenges in balancing user convenience with data privacy and security in AI assistants. How Meta will alter the feature to safeguard user information remains to be seen, but the incident emphasizes the need for cautious design where shared AI-generated content respects confidentiality and resists unauthorized access via public internet records.