MENLO PARK, CALIFORNIA / RankWire.AI / – Meta has put a stop to the sharing of chat links within its Muse AI assistant following the discovery by cybersecurity researchers that thousands of shared links could be retrieved through publicly accessible internet records. These links potentially exposed confidential personal and corporate data. Meta informed Ars Technica on Tuesday, October 20, that it had paused this feature while reviewing the findings. The company indicated it is also contemplating modifications to how shared links are made discoverable.
Security firm Claroc identified over 6,800 shared Muse links by analyzing passive DNS records. Such databases log domain-name queries and can render web addresses searchable. The figure represents discovered link addresses, not the total number of sensitive conversations. Claroc published its research on October 18, noting it first reported the issue to Meta’s bug bounty program on October 8, including steps to reproduce and example shares.
Researchers accessed several hundred conversations, with nearly all active shares loading without requiring a Meta account. Some of these contained sensitive details, such as an internal revenue review from a marketing firm and a small-business loan application. Others included a bank fraud alert with transaction amounts and partial account details. Additionally, confidential interview notes from executive searches and an employee vacation calendar were found. Many other shared chats consisted of routine content.
Link architecture made chats discoverable
The vulnerability originated from how Muse generated its share links. Each URL combined the chat’s title with a randomly generated identifier within the hostname. This identifier was sufficiently long to prevent guessing. However, when a link was accessed, it triggered a DNS lookup that passive DNS services could log and make searchable. As a result, others could locate these addresses without direct access and view the conversations without signing in.
Muse requested that search engines and AI crawlers exclude its share pages from indexing, but this did not prevent access via DNS records. The security site urlscan.io stored more than 75 saved scans of Muse share pages. Claroc stated it found no evidence of misuse, nor did it contact users impacted, and it remains unclear if others discovered the links. Meta’s official statement did not specify if affected users would be notified.
Sharing functionality temporarily disabled after review
On October 20, Claroc re-evaluated 40 links that had previously loaded without user accounts. All of these now returned errors, with content-serving addresses no longer functioning. Despite this, conversation titles still appeared in passive DNS histories. Meta launched Muse in the United States on September 8, targeting users aged 18 and above. The assistant can connect with users’ email, calendar, payment, and health applications. In April, Meta unveiled its Muse Spark model, developed in its superintelligence lab.
This incident underscores increasing concerns over data privacy in AI tools and the dangers of sharing conversations containing sensitive information. Similar issues arose in 2025, when shared ChatGPT dialogues appeared in Google search results. Meta has also faced legal challenges concerning user safety, including an order to pay $567 million following a New Mexico court ruling in August. The company has not announced a timeline for the return of Muse sharing features.