Claroc Researchers Expose Thousands of Sensitive Meta Muse Conversations via Single Query
Security researchers from Claroc found that a single passive DNS query reveals thousands of publicly accessible links containing private business and personal data shared via Meta’s new AI agent.
Key points
- Claroc researchers enumerated 6,800 shared conversation links from Meta Muse using one query.
- Exposed data includes loan applications, revenue reviews, vacation calendars, and fraud alerts.
- The vulnerability stems from URL construction design, despite Meta’s privacy-focused marketing.
Security researchers from Claroc reported on Monday that Meta Muse, the company’s new general-purpose AI agent, exposes sensitive personal and business information when users share conversations through web links. The firm stated that a single query allowed them to enumerate thousands of these links, revealing content that was clearly not intended for public access.
The researchers were able to identify 6,800 links created by Muse users for sharing conversations. After opening several hundred of these URLs, they discovered highly sensitive data. This included a candidate’s notes for a confidential executive search call, a national marketing company’s internal revenue review marked for its CEO, and a small-business owner’s loan application with the requested amount.
Context and Background
Other exposed documents included a to-do list featuring a bank fraud alert with a partial account number and transaction details. Researchers also found a real-estate investor’s internal performance report listing property addresses, exact returns, income, and mortgage figures. Additionally, a workplace’s full-year staff vacation calendar was accessible, listing dozens of employees against their scheduled weeks away.
Meta CEO Mark Zuckerberg stated last month that Muse was “built from the ground up for privacy and security.” The service allows users to share searches, conversations, and images via unique URLs, a feature common to many online platforms. Typically, these URLs contain long, unguessable text sequences to prevent unauthorized access. However, a central design decision in how Muse constructs these URLs has made them easy to discover.
Who Is Affected
According to Claroc, the easiest method to discover these URLs is to perform a passive DNS search through commercial services. Passive DNS records store historical resolution data for domain names to IP addresses, as well as nameserver and mail-server records. These systems collect DNS responses from sensors at recursive resolvers or other infrastructure, creating records when monitoring systems observe responses.
The researchers noted that the ease of access was surprising. “If you had asked us beforehand, we would have guessed that gathering these would take real effort,” they wrote in their report. “It took one query.” This suggests that the current implementation of the sharing feature fails to adequately obscure the existence of shared content from external enumeration techniques.
What should you do?
No fix or update for this issue has been confirmed by Meta yet. Users who have shared conversations via Muse links should assume the content is publicly accessible. Organizations should review any sensitive data shared through the platform and consider the exposure risk until a vendor patch is provided.
What to do and how to stay safe: Meta
- Audit any links shared via Meta Muse to determine if they contain sensitive personal or business information.
- Monitor passive DNS records for your organization’s domains to detect unexpected URL enumeration or exposure.
- Restrict the use of public sharing features for confidential communications until the vendor confirms a security fix.
- Educate employees on the risks of sharing sensitive data through AI agent platforms with public link capabilities.
Step-by-step guide: Mobile Device Security: How to Protect Data Beyond Passwords
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
How many sensitive links did researchers find?
Claroc researchers enumerated 6,800 links created by Muse users for sharing conversations.
What kind of data was exposed?
Exposed data included loan applications, revenue reviews, vacation calendars, fraud alerts, and real-estate performance reports.
Is there a patch for this vulnerability?
No fix has been confirmed yet by Meta regarding the URL enumeration issue.
