MENLO PARK, CALIFORNIA / RankWire.AI / – Meta has temporarily suspended the sharing of chat links within its Muse AI assistant following the discovery that thousands of shared conversations could be located through public internet records. These links risked revealing sensitive personal and corporate data. On Tuesday, October 20, Meta informed Ars Technica that it had paused this feature to assess the findings. The company also indicated it is exploring modifications to how shared links are detected and accessed.
The security firm Claroc uncovered over 6,800 shared Muse links via passive DNS records, which log domain lookups and allow web addresses to be searchable. This number reflects the total discovered link addresses, not necessarily the count of sensitive chats. Claroc released its research on October 18, revealing that it first alerted Meta’s bug bounty program on October 8, providing reproduction steps and sample shares.
A few hundred conversations were accessed by researchers, many of which loaded without requiring a Meta account. Among the exposed content were confidential details like an internal revenue review from a marketing firm, a small-business loan application, and a bank fraud alert showing a transaction amount along with partial account information. Researchers also found sensitive documents such as confidential interview notes for an executive search and an employee vacation calendar. Meanwhile, many other shared chats included routine or non-sensitive material.
Design flaw made chats publicly discoverable
The vulnerability originated from how Muse generated its share URLs. Each link incorporated the conversation’s title combined with a randomly assigned identifier within the hostname. This identifier was sufficiently long to prevent guessing. However, clicking on the link caused a DNS lookup, which passive DNS services could record and index. This meant that others could locate these addresses via DNS records without direct access, and view conversations without needing to sign in.
Despite Muse requesting search engines and AI crawlers not to index its share pages, this did not prevent access via DNS. The security site urlscan.io stored over 75 saved scans of Muse share pages. Claroc reported no evidence of misuse so far and did not contact the users affected, nor could they determine if others had discovered the links. Meta’s statement did not specify whether users impacted by this issue would be notified.
Sharing capabilities disabled following review
On October 20, Claroc retested 40 shares that had previously been accessible without an account. None of these loaded, and the associated addresses returned errors. However, the conversation titles remain stored passively in DNS history. Meta launched Muse in the United States on September 8, targeting users aged 18 and older. The assistant integrates with users’ email, calendar, health, and payment apps. Additionally, Meta introduced its Muse Spark model in April, developed by its superintelligence lab.
This incident underscores increasing concerns regarding data privacy in AI-driven applications, especially related to the risks of sharing sensitive information. Similar issues arose in 2025 when shared ChatGPT conversations appeared in Google search results. Meta also faces legal challenges concerning user safety, such as the $567 million court-ordered payment following a ruling in New Mexico in August. The company has not announced when the sharing feature for Muse will be reinstated.