Meta Halts Key Muse AI Agent Feature Over Privacy Concerns
Meta has temporarily suspended a conversation-sharing feature in its Muse AI assistant after cybersecurity researchers discovered that thousands of shared chat links could be found through public internet records. The company said it was reviewing the findings and considering changes to how the links could be discovered. The incident has raised questions about privacy protections in AI assistants, particularly tools designed to handle personal information, business documents, and other sensitive tasks on users’ behalf.
Researchers Discover Thousands of Shared Chat Links
Cybersecurity firm Claroc reportedly identified more than 6,800 shared Muse conversation links by examining passive DNS records, which preserve information about domain-name lookups. Researchers said the links could be discovered without being directly shared with them. The issue involved the way Muse generated web addresses for shared conversations. According to the researchers, the URL structure included conversation titles and identifiers, while the associated records could make the links discoverable through historical DNS data. The discovery highlighted how a feature intended to make conversations easier to share could also create unintended privacy risks.
Sensitive Personal and Business Information at Risk
Researchers said they were able to access conversations containing potentially sensitive information, including business revenue reviews, loan applications, bank fraud alerts with partial account details, job interview notes, and internal company documents. The findings did not mean that every Muse conversation was publicly accessible. The reported exposure concerned shared conversations whose links could be discovered through public records. Nevertheless, the incident demonstrated that information users choose to share can remain vulnerable if the mechanisms protecting those links are insufficient.
Meta Suspends the Sharing Feature
Meta confirmed that it had paused the conversation-sharing functionality while reviewing the researchers’ findings and considering changes to how shared links could be discovered. After the suspension, Claroc reported that a sample of links it had previously accessed was no longer available. The company’s response was a temporary suspension rather than an announcement that the entire Muse AI service had been shut down. The reported action specifically targeted the conversation-sharing feature while the review continued.
What Is Meta Muse AI?
Muse is Meta’s personal AI agent, introduced in September 2026. It is designed to help users carry out tasks such as managing information, working with connected services, shopping, and organizing activities. Meta says the system can operate through a dedicated cloud-based virtual machine and work with services users choose to connect. Because AI agents can interact with personal accounts and handle information on behalf of users, privacy and security are particularly important. A weakness in a sharing feature can create risks even when the underlying service has other safeguards in place.
Privacy Concerns Add to Scrutiny of AI Agents
The latest incident comes amid a wider debate about how AI agents access, process, and share sensitive information. Unlike conventional chatbots, agentic systems can perform tasks across connected applications, making permissions, user oversight, and secure data handling especially important. Meta has previously said that Muse includes controls for managing connected services and permissions. However, the shared-link incident illustrates that privacy also depends on how generated content is published, linked, and made accessible outside the user’s account.
What Happens Next?
Meta has said it is reviewing the issue and considering changes to the way Muse conversation links can be discovered. The company had not announced a confirmed date for restoring the feature in the reports available on October 23, 2026. Users should exercise caution when sharing AI conversations containing confidential business information, financial details, or personal data. Until further details are provided, the incident serves as a reminder to review sharing settings and avoid including sensitive information in content intended to be shared publicly.
