MENLO PARK, CALIFORNIA / RankWire.AI / – Meta has temporarily halted the sharing feature of its Muse AI assistant following security experts’ discovery that thousands of shared chat links could be found in public internet records. This exposure risked revealing sensitive personal and business information. On Tuesday, October 20, Meta announced it had paused the feature to investigate the incident further and is considering modifications to how shared links are accessible.
The discovery was made by security firm Claroc, which identified over 6,800 Muse share links within passive DNS logs. These logs record Domain Name System lookups, making website addresses accessible to anyone who searches the data. Claroc reported that it informed Meta via its bug bounty program on October 8 and published its findings on October 18. The figure of 6,800 pertains to found addresses, not sensitive conversation content.
Researchers from Claroc examined a few hundred of these shared conversations, nearly all of which could be accessed without login credentials. Included among them was a small-business loan application with partial account numbers, a note regarding a bank fraud alert, a marketing firm’s revenue report prepared for its CEO, and a real-estate investor’s mortgage and return analysis. Confidential interview prep notes also appeared in the shared data.
Chat titles exposed through hostname data
The core issue stemmed from Muse’s address design for sharing chats. Each link contained the conversation title and a randomly generated identifier within the hostname. Although this identifier was not easily guessable, visiting a link triggered a DNS lookup that passive DNS services recorded. This meant that even users who never received a link could find it through those logs. The shared chats could be accessed without requiring user authentication.
While Muse’s settings instructed search engines and AI bots to avoid its shared pages, researchers did not need search engines to locate them. Over 75 cached scans of Muse share pages were stored on urlscan.io. Claroc stated there was no evidence of misuse of the data and did not reach out to affected users. It also could not determine if others had discovered the links. Meta’s statement did not specify if users would be notified about the exposure.
Pause initiated as Meta assesses security findings
When Claroc re-verified 40 previously accessible shares on October 20, none loaded, and their addresses returned errors. The conversation titles, however, remain visible in passive DNS history. Muse officially launched in the U.S. on September 8, targeting adults aged 18 and over. It links with users’ email, calendar, payment, and health apps. Meta introduced its Muse Spark model, created by its superintelligence laboratory, in April.
This incident underscores ongoing privacy concerns surrounding AI tools and the dangers of sharing chats containing confidential data. In 2025, shared ChatGPT conversations appeared in Google search results, prompting OpenAI to eliminate the option that allowed indexing. Additionally, in August, Meta was ordered to fund teen mental health initiatives in New Mexico. Meta has yet to announce when Muse’s sharing feature will return or how the link system might be altered.