MENLO PARK, CALIFORNIA / RankWire.AI / – Meta has halted the sharing of conversations within its Muse AI assistant after cybersecurity experts uncovered that thousands of shared chat links could be traced through publicly accessible internet records. These links potentially revealed sensitive personal and corporate data. Meta informed Ars Technica on Tuesday, October 20, that it has paused this feature while it examines the findings. The company also indicated it is contemplating modifications to how shared links might be discovered in the future.
The security company Claroc identified over 6,800 shared Muse links via passive DNS records. These records log domain-name lookups, making URLs searchable. The reported figure pertains to discovered link addresses, not the total number of sensitive conversations. Claroc published its findings on October 18, stating it initially reported the issue to Meta’s bug bounty program on October 8, providing reproduction steps and sample shares.
Several hundred conversations were accessed by researchers, with nearly all live shares opening without requiring a Meta account. Some contained sensitive data, including an internal revenue review for a marketing firm and a small-business loan application. Others displayed a bank fraud alert along with a transaction amount and partial account information. Confidential notes from executive interviews and a workplace vacation calendar listing multiple employees were also uncovered. Many additional shared chats consisted of routine content.
Discovery due to link structure
The vulnerability originated from how Muse generated its share URLs. Each link combined the conversation’s title with a randomly generated identifier within the hostname. This identifier was sufficiently long to prevent guessing. However, clicking on a link initiated a DNS lookup, which passive DNS services could log and index. Consequently, others could locate these addresses without direct access and view chats without signing in.
Although Muse requested search engines and AI crawlers not to index its shared pages, this did not prevent access through DNS records. The security site urlscan.io stored more than 75 saved scans of Muse share pages. Claroc stated it did not find evidence of misuse. The firm did not contact users affected and was unable to determine if others had discovered the links. Meta’s official statement did not specify whether affected users would be notified.
Link sharing disabled following review
On October 20, Claroc re-evaluated 40 shared links that previously opened without a user account. All of these links now returned errors, and their URLs no longer served content. However, conversation titles still remain in passive DNS records. Meta introduced Muse in the United States on September 8, targeting users aged 18 and older. The assistant integrates with users’ email, calendar, payment, and health applications. In April, Meta announced the Muse Spark model from its superintelligence laboratory.
This incident emphasizes the increasing concerns about data privacy in AI tools and the dangers of sharing confidential conversation content. Similar issues arose in 2025 when shared ChatGPT chats appeared in Google search results. Meta has also faced legal challenges related to user safety, including a court-ordered payment of $567 million following a New Mexico ruling in August. The company has not announced when Muse chat sharing will resume.