MENLO PARK, CALIFORNIA / RankWire.AI / – In response to security concerns, Meta has temporarily disabled the sharing feature for its Muse AI assistant following findings by security researchers indicating that thousands of shared chat links could be accessed via public internet records. This exposure posed the risk of revealing sensitive personal and corporate information. On Tuesday, October 20, Meta announced that the feature was on hold while it evaluates the situation and considers potential adjustments to how shared links are discovered.
The discovery was made by security firm Claroc, which identified over 6,800 Muse share links within passive DNS data. These DNS records record Domain Name System lookups and can expose website addresses to anyone examining the data. Claroc reported that it notified Meta through its bug bounty program on October 8 and published its findings on October 18. It’s important to note that the 6,800 figure reflects identified addresses, not the contents of any conversations.
A review of a few hundred of these shared conversations by Claroc’s team revealed that nearly all of the live shares could be accessed without requiring a login. Among the data uncovered were a small-business owner’s loan application containing part of an account number, as well as a note referencing a bank fraud alert. Other examples included a marketing firm’s revenue review prepared for its CEO and a real-estate investor’s report detailing mortgage information and investment returns. Confidential interview notes also appeared to be publicly accessible.
Hostnames exposed chat titles
The root of the issue was linked to how Muse’s sharing URLs were structured. Each link included a hostname that combined the conversation’s title with a random identifier. While this identifier was not easily guessable, visiting a link triggered a DNS lookup that passive DNS services could log. Consequently, individuals who had not received the link directly could still find it through these databases. Such shared conversations could then be accessed without any login prompt.
Despite Muse’s configuration instructing search engines and AI crawlers to ignore its shared pages, researchers found no need for search engines to locate them. Over 75 saved scans of Muse share pages were stored on urlscan.io. Claroc clarified that it found no evidence of misuse of the data. The firm did not contact the users affected and could not determine whether others had discovered the links. Meta’s official statement did not specify whether users would be informed about the breach.
Feature suspension underway as Meta reviews findings
On October 20, Claroc revisited 40 of the previously accessible shares, and none of the links opened, instead returning errors. The conversation titles remain visible in passive DNS logs. Muse was launched in the United States on September 8, targeting users aged 18 and above. It offers integrations with email, calendar, payment, and health applications. Additionally, Meta introduced its Muse Spark model, created by its superintelligence lab, in April.
This incident underscores ongoing privacy concerns surrounding AI chat tools and the potential risks associated with sharing confidential conversations. Notably, in 2025, shared ChatGPT conversations appeared in Google search results, which led OpenAI to remove the feature that made these chats indexable. Furthermore, in August, Meta was ordered to fund teen mental health programs in New Mexico. Meta has not yet announced when Muse’s sharing capabilities will be reinstated or how the links will be handled going forward.