MENLO PARK, CALIFORNIA / RankWire.AI / – Meta has temporarily halted the sharing feature in its Muse AI assistant after security experts revealed that thousands of shared chat URLs could be accessed through public internet records. This exposure risked revealing sensitive personal and corporate information. On Tuesday, October 20, Meta announced it had paused the feature while reviewing the findings. The company is considering modifications to how shared links are discovered.
The discovery was made by security firm Claroc, which identified more than 6,800 Muse share links within passive DNS records. These records log Domain Name System lookups, making website addresses accessible to anyone inspecting the data. Claroc reported that it informed Meta via its bug bounty program on October 8 and published its research on October 18. The 6,800 figure reflects the number of discovered addresses, not the number of sensitive conversations.
Researchers at Claroc examined several hundred shared conversations, almost all of which were accessible without login credentials. Among the shared data were a small-business owner’s loan application containing part of an account number, along with a note about a bank fraud alert. Additional examples included a marketing firm’s revenue report prepared for its CEO, as well as a real-estate investor’s report on mortgages and investment returns. Confidential interview prep notes also appeared in some shared links.
Hostnames exposed chat titles
The core issue stemmed from the design of Muse’s sharing URLs. Each URL included the conversation title and a unique, randomly generated identifier within the hostname. While the identifier was not straightforward to guess, visiting a link triggered a DNS lookup that passive DNS services could log. This allowed anyone, even those who had not received the link directly, to locate it in these databases. The shared conversations could then be accessed without requiring a login.
Although Muse’s settings instructed search engines and AI crawlers to avoid indexing its share pages, researchers did not depend on search engines to locate them. More than 75 saved scans of Muse share pages were stored on urlscan.io. Claroc clarified that there is no evidence indicating that anyone misused the data. The firm did not reach out to users affected and could not determine whether others had discovered the shared links. Meta’s official statement did not specify whether users would be notified.
Feature put on hold during review process
On October 20, Claroc re-examined 40 previously accessible shared links, all of which now returned errors and failed to load. The conversation titles still appear in passive DNS records. Muse was launched in the United States on September 8, targeting users aged 18 and above. It integrates with users’ email, calendar, payment, and health applications. In April, Meta introduced its Muse Spark model, created by its superintelligence lab.
This incident underscores ongoing privacy concerns surrounding AI tools and the dangers associated with sharing chats containing confidential data. In 2025, shared ChatGPT conversations appeared in Google search results, prompting OpenAI to eliminate the feature that allowed indexing. Additionally, in August, Meta was ordered to fund teen mental health programs in New Mexico. Meta has yet to announce when Muse’s sharing capabilities will be restored or how the link sharing system might be altered.