Editorial illustration for Meta pauses Muse sharing after security firm uncovers AI chats on the web

Meta pauses Muse sharing after security firm uncovers AI...

By AI Daily Post Edited by Brian Petersen, Editor-in-Chief

October 20, 2026 • 3 min read

Meta shut off link sharing for its Muse AI assistant this week after a security firm found thousands of private conversations sitting out in the open on the public internet. Claroc, the firm behind the discovery, says it pulled more than 6,800 Muse share links by digging through public DNS records, no hacking required. The exposed chats weren't abstract test data either. Among them: a marketing company's revenue figures, a workplace calendar, and a loan application, the kind of material nobody intends to hand over when they hit "share" on an AI chat.

The problem traces back to how Meta built the sharing feature in the first place. Users could generate a link to a specific Muse conversation and send it to someone else, a fairly standard setup. But the structure of those URLs turned out to be the weak point, letting outsiders reconstruct and locate chats they were never meant to see.

Meta spokesperson Tracy Clayton confirmed the pause to The Verge, saying the company is reviewing Claroc's findings and looking at how to change the way these links get discovered. For now, the sharing function is offline while that review plays out.

By searching public DNS records, which serve as a directory for web traffic, Claroc was able to view hundreds of Muse conversations, including some containing a marketing company’s revenue review, a small business owner’s loan application, a workplace’s full-year staff calendar, and notes for a job candidate’s call with an AI lab.

Meta pauses Muse sharing after researchers find AI chats on the web - The Verge AI

Why this matters

Claroc's 6,800-link haul is a reminder that "shareable" and "public" are two very different promises, and Meta apparently blurred them. Shareable links indexed through public DNS records aren't a leak in the traditional sense, they're a design choice that nobody stress-tested against search crawlers. For developers bolting on a "share chat" button to any AI tool, this is the exact failure mode to check for before launch: are links guessable, are they indexed, do they expire.

For founders building on Meta's AI stack, the pause itself is the bigger signal. Shipping a feature, then yanking it after an outside security firm finds the problem rather than internal red-teaming, suggests the sharing mechanics weren't pressure-tested for exposure before release. Loan applications and payroll calendars sitting in public URLs isn't a hypothetical privacy risk, it's a demonstrated one, with a count attached.

Anyone using Muse to discuss anything resembling business or financial data should treat "share" as "publish" until Meta says otherwise, and frankly build their own products with that assumption baked in from day one.