Meta has halted the share link functionality for its AI assistant Muse following a security researcher discovery that share URLs included conversation titles in subdomains, making them easy to enumerate and exposing confidential personal and business information.
- Muse share URLs embed conversation titles in subdomains, enabling easy public discovery
- Claroc security researchers enumerated thousands of private shared links via passive DNS
- Meta paused share feature and launched a bug bounty review to address the flaw
What happened
Meta’s Muse AI assistant includes a feature allowing users to share conversations and other content via unique URLs. These URLs are meant to be secure, containing unguessable strings so only people with the link can view the shared material. However, a design flaw in how these links are constructed placed the conversation title within the subdomain portion of the URL rather than the path.
Security researchers at Claroc discovered this setup exposed those links in public passive DNS records, allowing them to enumerate about 6,800 share links and access sensitive, private user information not intended for public viewing. Meta responded by suspending the share feature while reviewing security improvements.
Why it matters
The exposure of personal and corporate data shows how critical URL design and DNS considerations are for protecting user privacy. Unlike typical share links that keep identifying details out of DNS-recorded domains, Muse’s approach unintentionally left a searchable trail of sensitive conversation topics and metadata.
With Muse rising rapidly in popularity for automating tasks like form filling and document creation, the reach of this vulnerability was broad and impactful, affecting individuals and businesses sharing confidential data via the assistant. The incident highlights ongoing challenges in securing AI-driven productivity tools.
What to watch next
Meta is currently evaluating fixes through its bug bounty program and has temporarily disabled Muse’s share links. Observers will be looking for how Meta addresses URL construction, potentially moving conversation metadata out of DNS-visible subdomains to prevent similar leaks.
The episode signals greater scrutiny over privacy and security controls in AI assistants, suggesting regulators and users alike will demand stronger protections and more transparent security practices as these tools become more integrated into sensitive workflows.