GoKawiilMeta disables Muse chat-sharing after researchers expose thousands of private links
Security firm Claroc found that Meta's Muse AI agent generated shareable chat URLs using the conversation's title as a subdomain, which were recorded in public DNS and discoverable via passive DNS searches. Claroc said it enumerated roughly 6,800 shared Muse links, exposing personal and business data including meeting schedules, a loan application, a bank fraud alert and internal company documents. Meta confirmed to Ars Technica that it has paused the sharing feature while reviewing the report.
GoKawiil's interpretation of the reporting above, not reported fact.
The incident shows how a seemingly minor design choice—embedding content-derived titles in public DNS records—can turn an AI assistant's convenience feature into a mass data-exposure risk. It also suggests that as agentic AI tools proliferate and handle sensitive personal and business tasks, companies may need to rethink basic infrastructure decisions, not just model safety, to avoid leaking private data at scale.
- Muse's shared chat links used predictable, title-based subdomains that were indexed in public DNS records.
- Claroc says it accessed about 6,800 exposed links containing sensitive personal and business information.
- Meta has paused the sharing feature and is reviewing how links can be discovered, though some previously found links reportedly no longer work.
Source: engadget.com — Anna Washenko, 2026-10-20
Published there as: “Meta suspends Muse chats sharing feature after it left private details exposed”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.