MENLO PARK, CALIFORNIA / RankWire.AI / – Meta has paused the sharing of conversations within its Muse AI assistant following discoveries by cybersecurity experts indicating that thousands of shared chat links could be located through publicly accessible internet records. These links potentially revealed sensitive personal and corporate data. On Tuesday, October 20, Meta informed Ars Technica that it had temporarily disabled this feature while examining the findings. The company also indicated that it is contemplating modifications to how shared links are detected.
Security firm Claroc identified over 6,800 shared Muse links via passive DNS records. These records track domain name lookups, making web addresses searchable. The reported number reflects the discovered link addresses, not the total number of sensitive conversations involved. Claroc published its research on October 18, stating that it initially reported the issue to Meta’s bug bounty program on October 8, providing reproduction steps and example links.
Several hundred conversations were accessed by researchers, with nearly all active shares opening without a Meta account. Some of these contained highly sensitive data, such as an internal revenue review from a marketing firm and a small-business loan application. Others displayed a bank fraud alert, including transaction amounts and partial account details. Researchers also uncovered confidential interview notes for executive searches and an employee vacation calendar. Many other shared conversations consisted of routine content.
Chat discoverability caused by link structure
The vulnerability originated from how Muse generated its sharing URLs. Each link incorporated the conversation’s title along with a randomly generated identifier embedded within the hostname. This identifier was sufficiently long to prevent guessing. However, when a link was opened, it triggered a DNS lookup, which passive DNS services could log and make searchable. This meant that third parties could locate the links without direct access and view the chats without needing to sign in.
Although Muse instructed search engines and AI web crawlers not to index its shared pages, this did not prevent access through DNS records. The security website urlscan.io stored more than 75 saved scans of Muse share links. Claroc stated that no evidence of misuse was found. The firm did not notify affected users and was unable to determine if others had discovered the links. Meta’s official statement did not specify whether affected users would be notified.
Sharing links deactivated following review
On October 20, Claroc re-evaluated 40 shared links that previously loaded without a Meta account. All of these returned errors, and the content-serving addresses no longer functioned. However, the conversation titles still appeared in passive DNS records. Meta introduced Muse in the United States on September 8, targeting users aged 18 and older. The assistant can connect with users’ email, calendar, payment, and health applications. In April, Meta announced its Muse Spark model developed within its superintelligence research division.
This incident underscores the increasing anxiety over data privacy issues in AI-based tools and the dangers associated with sharing conversations that include confidential information. Similar vulnerabilities emerged in 2025 when shared ChatGPT interactions appeared in Google search results. Additionally, Meta has faced legal challenges related to user safety, including a court-mandated payment of $567 million following a New Mexico court ruling in August. The company has not yet announced when Muse’s sharing capabilities will be reinstated.