Meta pauses Muse chat sharing after 6,800 links surface in DNS records

Meta has paused chat sharing in its Muse assistant after the security firm Claroc reported finding more than 6,800 share links in public DNS records, The Verge reported on October 20. Meta says it is reviewing the findings.

How share links ended up in DNS records

Muse lets users turn a conversation into a page that anyone with the link can open. According to Claroc’s write-up, the page address is built from a short title describing the chat, followed by a random ID. Claroc says it searched a commercial passive-DNS service for hostnames under Muse’s content domain and did not guess or brute-force anything. After filtering out noise from wildcard DNS entries, it counted more than 6,800 share hostnames.

Claroc opened only a few hundred of them and says almost every live one loaded without an account. Its argument is that DNS logs the full hostname on the first lookup, so the random ID gives no protection once the title is in the address. The Verge quotes Claroc as saying a hostname pulled from DNS can rebuild the link.

Among the pages, Claroc describes categories including a marketing company’s revenue review, a loan application, a staff vacation calendar and a candidate’s notes for a call with an AI lab. It says most of the pages it opened were harmless. A separate public scan database, urlscan.io, also held more than 75 saved scans of Muse share pages, according to Claroc.

What Meta has said so far

Muse spokesperson Tracy Clayton told The Verge that people can share specific Muse artifacts, creating a page reachable by anyone with the link. Meta has “paused this functionality while we review this report,” she said, and will consider changes to how the links can be discovered.

Claroc says it rechecked 40 shares on October 20 that had opened without an account earlier in the month. None opened any more. The hostnames and their titles, however, remain in passive-DNS history.

A familiar privacy pattern with a new twist

The exposure is not the result of a break-in. Claroc says Meta’s own help page for Meta AI states that people with the link can see the conversation, that links do not expire and that they can be passed on. Claroc notes that this page covers Meta AI in general rather than Muse specifically. The problem it describes is that a link meant for one colleague was effectively listed in a public directory.

Earlier cases involved search engines. The Verge notes that shared chats from ChatGPT, Claude and Grok have also ended up in Google Search. The security write-up that Unite.AI summarized puts the number of indexed Grok conversations at more than 370,000. Muse differs because the discovery channel was DNS, not a search index, so telling crawlers to stay away would not have helped.

The timeline matters too. Muse launched on September 8, 2026, according to Claroc, which reported the issue to Meta’s bug bounty program on October 8. Meta’s pause came on October 20, after Ars Technica asked about the findings, Claroc says. Meta’s response to The Verge does not mention the bug bounty report.

What the findings do not establish

Claroc is a security vendor, and its post promotes its own product, so the account has not been independently verified. The firm says it examined one discovery channel, one hostname pattern, one product and a short window, so 6,800 is not a measure of total exposure. It also cannot say how long any share was reachable or whether others found the pages first. Claroc says it found no evidence of misuse and did not contact the people or companies involved.

Meta has not said when sharing will return or what will change. For Muse users, the practical point is that links already created may sit in DNS history even though the pages no longer load. Anything pasted into a chat that was shared before the pause is worth treating as potentially seen by others.

The Muse story comes at a busy moment for Meta’s assistant. CNBC reported on October 15 that Apple and Meta are teaming up to run Muse on Mac computers. It also arrives as AI security is drawing wider attention, such as a reported case of AI agents run against scores of websites.