Meta pauses Muse chat sharing after 6,800 links surface online

By Marcus Bennett

Meta has paused a Muse feature that allowed users to share conversations with the AI assistant, after security researchers discovered more than 6,800 Muse links publicly available on the web. The count comes from Claroc, a security firm, which documented the findings in a report published this week.

The exposure matters because shared AI conversations can contain personal and business information, and the links were reachable by anyone who knew where to look. Meta confirmed the suspension of the feature to The Verge.

What did the researchers actually find?

Researchers at Claroc say they located more than 6,800 Muse links through public DNS records (the distributed lookup system that translates domain names into network addresses). DNS queries are routinely logged and exposed through passive DNS services, which archive lookup activity across the internet.

That mechanism allowed the researchers to assemble a corpus of shared conversation links without breaching any Meta system. Ars Technica reported the findings earlier, noting that the shared Muse conversations could expose users' personal and business secrets.

Claroc published its full analysis, titled "GhostShares," on the firm's blog this week.

What is Meta's response?

Muse spokesperson Tracy Clayton confirmed the feature suspension in an emailed statement to The Verge: "People have the ability to share specific Muse artifacts, which creates a page that can be accessed by those with the link. However, we've paused this functionality while we review this report and consider changes to how these links can be dis…"

The statement was truncated in the published excerpt. The key fact stands regardless: sharing is off while Meta reviews the Claroc report.

Meta has not stated when the feature might return.

Why does this matter for AI assistants?

The incident turns on a design pattern that has repeatedly caused privacy problems across the software industry: the "anyone with the link can view" sharing model. When conversation links are unguessable but permanent, and when their existence leaks through infrastructure like DNS logs, the practical privacy guarantee weakens considerably.

For AI assistants, the stakes are higher than for ordinary document sharing. Conversations with an assistant can include:

• Personal disclosures typed in the expectation of a private exchange
• Business information pasted into prompts for summarization or drafting
• Multi-turn context that accumulates sensitive detail over a session

More than 6,800 links appeared in public DNS records, according to Claroc. Each link represents a conversation page that, in Meta's own description, "can be accessed by those with the link."

What happens next?

Meta says it is reviewing the report and considering changes to how shared Muse links work. The company has not detailed what those changes would be, or whether previously shared links remain accessible during the pause.

The Claroc report and Meta's decision to suspend the feature will add pressure on AI assistant makers to treat shareable conversation links as sensitive infrastructure rather than a convenience feature.