Meta's Muse Exposes Private Conversations Through Predictable Share Links
Security researchers discovered a design flaw in Meta's Muse AI agent that allows attackers to enumerate and access thousands of private user conversations via predictable URLs.
Last verified:
BLUF
Meta’s Muse AI assistant contains a URL-design flaw that allows researchers to enumerate and access thousands of private user conversations. According to Wired AI, security firm Claroc discovered that Muse’s share links use predictable patterns, enabling attackers to discover 6,800+ links through a single passive DNS query and access sensitive information including interview notes, financial records, and employee data. Meta has suspended the sharing feature pending a redesign.
Muse’s Predictable URL Construction Enables Mass Enumeration
According to Wired AI, Meta’s Muse appears to rely on long, random token sequences in share links—the same approach most online services use to prevent unauthorized access. However, Claroc researchers identified a critical design flaw: the underlying structure of how Muse constructs these URLs follows patterns discoverable through passive DNS searches, a commercial service that logs historical domain-name resolutions over time.
The attack requires minimal effort. Wired AI reports that Claroc performed a single passive DNS query and successfully enumerated 6,800 share links created by Muse users. After opening a sample of these links, researchers accessed conversations clearly intended as private, including sensitive business and personal documents.
What Private Data Became Accessible
The exposed conversations span multiple categories of sensitive information, according to Wired AI’s reporting. Researchers found a candidate’s interview-preparation notes for a confidential executive-search call, a national marketing firm’s internal revenue review marked for CEO review, and a small-business owner’s loan application showing the requested amount. Additional disclosures included a workplace’s complete year-long employee vacation calendar listing dozens of staff members and their scheduled time off, a real-estate investor’s internal performance report containing two property addresses with exact return and mortgage figures, and a to-do list referencing a bank fraud alert with partial account numbers and transaction details.
Meta’s Response and Feature Suspension
In a statement to Wired AI, Meta acknowledged receiving the report through its bug bounty program and stated it is suspending Muse’s sharing functionality while the company reviews the design. Meta said: “We’ve paused this functionality while we review this report and consider changes to how these links can be discovered.”
According to Wired AI, Claroc reported the vulnerability on October 8, 2026, through Meta’s official bug bounty channel.
Why This Matters
This flaw contradicts Meta CEO Mark Zuckerberg’s recent claim that Muse was “built from the ground up for privacy and security.” For enterprises and individuals using Muse to share AI-generated analyses, the vulnerability means any conversation marked as shareable via link—even if intended for a single recipient—could be systematically discovered and accessed by a threat actor with basic DNS-query knowledge.
The incident highlights a recurring pattern in AI product security: sharing features designed for convenience often outpace the security review cycle. Teams evaluating Muse for sensitive workflows (hiring, financial analysis, legal research) should treat the paused feature as a signal to avoid sharing links containing identifying information until Meta provides technical details on the redesigned sharing mechanism. The bug bounty disclosure suggests Meta’s security processes caught the issue through external research rather than pre-launch testing—a concerning gap for a product marketed on privacy grounds.
Frequently Asked Questions
How did researchers discover the vulnerability?
According to Wired AI, Claroc researchers used passive DNS searches—a commercial service that logs historical domain-name resolutions—to systematically discover predictable share-link patterns used by Muse.
What types of sensitive data were exposed?
Wired AI reports that researchers found executive-search call notes with internal contact information, internal revenue reviews, loan applications with requested amounts, workplace vacation calendars, and real-estate performance reports with property addresses and financial figures.
Did Meta acknowledge the issue?
Yes. According to Wired AI, Meta stated it opened its bug bounty program for security research and has paused the Muse sharing feature while reviewing the report and considering design changes.
When was the vulnerability reported?
Wired AI reports that Claroc disclosed the findings through Meta's bug bounty program on October 8, 2026.