Industry

Meta Pauses Muse Share Feature After 6,800+ Conversations Exposed via DNS Enumeration

Security researchers discovered Meta's AI chat-sharing links were predictable enough to enumerate publicly, exposing sensitive business and personal data.

Last verified:

According to The Verge, Meta has suspended its Muse conversation-sharing feature following a disclosure by security firm Claroc that over 6,800 AI chat links were discoverable through public DNS queries. The vulnerability allowed researchers to reconstruct and access shared conversations containing sensitive business and personal information—including revenue data, loan applications, and workplace calendars—without requiring explicit link sharing from users.

How Claroc Discovered the Vulnerability

The attack exploited Meta’s link-generation scheme for Muse shares. According to Claroc’s report as cited by The Verge, each shared conversation generated a URL constructed from three components: the conversation’s title, a brief description of its contents, and a random identifier. Because DNS records (public directories that route web traffic) retained these predictable elements, Claroc could query DNS logs to systematically reconstruct and enumerate valid links.

Muse spokesperson Tracy Clayton confirmed the pause to The Verge in a statement, saying the company was “reviewing this report and considering changes to how these links can be discovered.” The feature’s design flaw was architectural rather than accidental—the hostname itself revealed the page’s contents before opening it, and since the link carried the same identifying information, DNS enumeration alone was sufficient to rebuild functional share URLs.

Exposed Data Categories

The conversations accessed by Claroc’s research touched multiple threat categories. According to The Verge’s reporting, discovered chats included corporate financial reviews, personal loan applications, employment records, and recruitment discussions with AI researchers. The variety suggests the vulnerability affected users across industries and use cases, not a single at-risk segment.

Pattern Across AI Platforms

This incident is not isolated within the AI industry. The Verge notes that conversations shared via OpenAI’s ChatGPT, Anthropic’s Claude, and xAI’s Grok have similarly appeared in Google Search results—indicating a systemic issue where AI vendors ship sharing features without adequately preventing indexing or enumeration.

Why This Matters

The pause signals growing friction between AI vendors’ desire to enable collaboration through shareable links and the security burden of preventing unintended exposure. For enterprise users of Muse—particularly those sharing financial, HR, or recruitment data—the incident underscores the risk of outsourcing sensitive conversations to AI platforms without end-to-end encryption or stricter access controls. Meta’s response will likely set expectations for how other vendors redesign sharing: moving from predictable, DNS-queryable URLs toward opaque tokens or cryptographic secrets. Until redesigned, users should assume any AI conversation link is discoverable and avoid sharing sensitive business information through platform-native sharing mechanisms.

Frequently Asked Questions

How did researchers find the exposed Muse conversations?

According to The Verge, security firm Claroc queried public DNS records—which log web traffic routing—to reconstruct Muse share links. Because Meta built links from predictable components (conversation title, short description, random ID), DNS queries alone revealed the link structure without needing the full URL.

What sensitive data was exposed?

According to Claroc's report cited by The Verge, exposed conversations included a marketing company's revenue review, a small business owner's loan application, a workplace's full-year staff calendar, and job candidate interview notes from an AI lab.

Is this the first AI chat-sharing privacy incident?

No. The Verge notes that conversations with OpenAI's ChatGPT, Anthropic's Claude, and xAI's Grok have similarly wound up indexed in Google Search, indicating a pattern of overly-public AI conversation sharing across vendors.

#ai-safety #privacy #meta #vulnerability-disclosure #dns-enumeration