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

Meta has paused a Muse feature that let users share AI conversations after researchers found more than 6,800 of those share links exposed publicly on the web. Security firm Claroc documented the findings in a report published this week, and Ars Technica first reported the story on Wednesday.

The numbers are the story. Claroc says it recovered the 6,800-plus Muse links not by breaking into anything, but by combing public DNS records — the internet's address book, which logs lookups that anyone can passively observe. Each link pointed to a page hosting a user's conversation with Meta's AI assistant. The report dubs the phenomenon "GhostShares": shared artifacts that users may have believed were private or at least obscure, sitting one URL away from anyone curious enough to look.

Meta's response came fast. The company confirmed it has suspended the sharing functionality while it reviews the report.

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

The statement cuts to the core of the problem. Links-as-permissions are only as private as the links themselves, and DNS traffic leaks link after link into publicly visible records.

What does the leak actually expose?

According to Claroc's report, the exposed links led to Muse conversation pages — chats that users had opted to share as artifacts. Ars Technica's coverage characterizes the findings as exposing users' personal and business secrets carried in those shared conversations.

The mechanics matter more than the count. Claroc didn't need credentials, exploits, or access to Meta's systems. Passive DNS monitoring — a technique security researchers use routinely — yielded the URLs. That means the exposure window isn't theoretical; anyone running similar infrastructure could have collected the same set of links before they were flagged.

For a company shipping an AI assistant to millions, that's an uncomfortable posture. Shared chats are a growth feature for AI products: they let users show off answers, collaborate, and effectively market the assistant for free. But each share creates a durable, addressable page, and durable pages get indexed, logged, and leaked.

Why did Meta move so quickly?

Meta says it paused the feature rather than patching quietly around it — an unusually direct response that suggests the company saw real exposure risk in the report. Tracy Clayton's statement leaves the door open to a redesign: the pause is "while we review this report and consider changes to how these links" work, implying the sharing feature could return with different link mechanics.

The episode lands at a moment when AI assistants increasingly hold sensitive material — business plans, personal confessions, code — and platforms push sharing as a feature. Claroc's 6,800-link haul demonstrates how quickly convenience features can collide with the ambient leakiness of internet infrastructure.

What happens next?

The sharing functionality stays off while Meta weighs changes. Watch for a return of the feature with hardened link design — expiring URLs, authentication gates, or scrambled identifiers — as Meta works to keep Muse's sharing growth engine running without leaving conversations discoverable through public DNS data.