Meta has paused a sharing feature for its Muse AI assistant after researchers found thousands of conversation links reachable on the open web. The company confirmed the change on Tuesday, shortly after a security firm published its findings.
Claroc, a security research company, discovered more than 6,800 Muse links through public DNS records. Those records act as a public directory for web traffic, and searching them revealed far more than any single user ever intended to expose.
“People have the ability to share specific Muse artifacts, which creates a page that can be accessed by those with the link,” Muse spokesperson Tracy Clayton said in a statement. “However, we’ve paused this functionality while we review this report and consider changes to how these links can be discovered.”
The way Meta built the links caused the exposure. Muse assembled each share address from the conversation’s title, a short phrase describing it, and then a random ID. Because the title described the contents, anyone scanning DNS records could recognise promising addresses at a glance.
Claroc reviewed hundreds of conversations before publishing its report. The exposed pages included a marketing company’s revenue review and a small business owner’s loan application. Researchers also found a workplace’s full-year staff calendar alongside notes for a job candidate’s call with an AI lab.
That mixture shows how much ordinary business now flows through an AI assistant. A link that looks harmless can expose revenue figures, hiring plans, or private financial details. None of those topics are unusual for a working conversation with an assistant.
The lesson echoes earlier failures in cloud document sharing. Public links that depend on obscurity collapse once the underlying addresses can be enumerated. Adding a readable title to a URL makes discovery dramatically easier than a random string alone.
Meta says it is weighing changes to how shared links can be discovered, but has not said when sharing returns. Muse conversations stay available inside the app while the sharing feature remains offline.
Anyone who already shared a Muse conversation should treat that link as exposed. Cached copies and search indexes can keep a page reachable long after the original is pulled.
The episode lands as AI assistants absorb more workplace tasks. Assistants now read calendars, draft documents, and hold context across long projects, which raises the cost of a single leaked conversation.
Security researchers expect similar findings at other AI companies. Share features ship quickly and usually get tested for function rather than discovery. The fix is not only a code change, but a review of what a link reveals before anyone clicks it.
Read the original report at The Verge.
Related reading: ChatGPT can read your Chrome tabs and fill out forms and how online adverts delivered zero-click spyware.
The Netherlands, for internationals
One email a week: the week's news for internationals in the Netherlands, new practical guides and what changed in the cities. No spam, unsubscribe any time.
