MENLO PARK, CALIFORNIA / RankWire.AI / – Meta has temporarily suspended the ability to share conversations via its Muse AI assistant after cybersecurity experts uncovered that thousands of shared chat links could be located through public internet records. These links may have inadvertently exposed sensitive personal and corporate data. On Tuesday, October 20, Meta informed Ars Technica that it has paused this feature while assessing the findings. The company is also exploring modifications to how shared links are accessible.
Security firm Claroc detected over 6,800 shared Muse links through passive DNS data. These records log domain-name lookups, making web addresses searchable. The figure reflects discovered link addresses, not the total number of sensitive chats. 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.
Researchers accessed several hundred conversations, with nearly all live shares opening without a Meta account. Some contained confidential information, including an internal revenue review from a marketing firm and a small-business loan application. Others revealed a bank fraud alert with transaction details and partial account info. Additionally, there were confidential interview notes for executive searches and a workplace vacation calendar listing dozens of employees. Many shared chats only involved routine content.
Link structure made chats discoverable
The security flaw originated from Muse’s approach to constructing shared link addresses. Each link combined the conversation’s title with a lengthy, randomly generated identifier within the hostname. While this identifier was complex enough to prevent guessing, opening a link triggered a DNS lookup. Passive DNS services could record these lookups, making the links searchable. Others could then locate these addresses without direct access and view chats without signing in.
Although Muse instructed search engines and AI crawlers not to index share pages, this did not prevent access via DNS records. The website urlscan.io stored more than 75 saved scans of Muse share pages. Claroc stated it found no evidence of misuse but did not contact the affected users or determine if others had discovered the links. Meta’s official statement did not specify whether users would be notified.
Sharing disabled following review
On October 20, Claroc re-examined 40 shares previously accessible without accounts, finding none loaded and their addresses returning errors. However, conversation titles remain visible in passive DNS logs. Meta launched Muse in the United States on September 8, targeting users aged 18 and older. The assistant integrates with email, calendar, payment, and health applications. In April, Meta introduced its Muse Spark model from its superintelligence laboratory.
This incident underscores the rising concerns over data privacy in AI tools and the risks involved when sharing sensitive information. Similar issues arose in 2025, when shared ChatGPT conversations appeared in Google search results. Meta also faces legal challenges, including a $567 million payout mandated by a New Mexico court in August. Meta has not indicated when chat sharing on Muse will resume.