MENLO PARK, CALIFORNIA / RankWire.AI / – Meta has paused the sharing of conversations in its Muse AI assistant following discoveries by cybersecurity experts. Researchers found that thousands of shared chat links could be located via publicly accessible internet records. These links may have revealed sensitive personal and corporate data. Meta informed Ars Technica on Tuesday, October 20, that the feature is currently suspended while it assesses the findings. The company is also considering modifications to how shared links are discoverable.
The security company Claroc uncovered over 6,800 shared Muse links through passive DNS data. These records log domain-name queries and allow searchability of web addresses. The reported figure refers to link addresses found, not the number of conversations with sensitive information. Claroc published its research on October 18. It stated that it first reported the problem to Meta’s bug bounty program on October 8, including steps to reproduce and example shares.
Researchers accessed several hundred conversations. Nearly all shared links loaded without a Meta account. Some contained confidential information, such as an internal revenue review from a marketing firm and a small-business loan application. Others displayed a bank fraud alert showing a transaction amount and partial account details. They also uncovered secret interview notes for executive searches and a workplace vacation calendar listing dozens of employees. Many other shared chats involved routine content.
Link structure made chats easily discoverable
The vulnerability came from how Muse generated its share URLs. Each link included the conversation’s title combined with a randomly assigned identifier in the hostname. This identifier was long enough to prevent guessing. However, opening a link caused a DNS lookup. Passive DNS services could record these lookups and make the addresses searchable. Others could then find the links without direct access and view chats without signing in.
Muse requested search engines and AI crawlers not to index its shared pages, but this did not prevent access via DNS records. The security site urlscan.io stored more than 75 scans of Muse share pages. Claroc stated it found no evidence of misuse. The firm did not contact affected users and could not determine if others discovered the links. Meta’s statement did not specify whether users would be notified.
Sharing links are now disabled pending review
On October 20, Claroc rechecked 40 shared links that had previously loaded without user accounts. All of them now returned errors, and the content addresses no longer served data. Conversation titles, however, remain in passive DNS history. Meta launched Muse in the United States on September 8. It is available to users 18 and older. The assistant connects with email, calendar, payment, and health apps. In April, Meta introduced its Muse Spark model from its superintelligence laboratory.
This event underscores increasing concerns over privacy in AI tools and the risks of sharing confidential conversations. Similar issues appeared in 2025, when shared ChatGPT chats appeared in Google search results. Meta has also faced legal challenges, including a $567 million penalty after a New Mexico court ruling in August. The company has not announced when Muse’s sharing features will be restored.