#dns

Researchers at security firm Claroc say they found 6,800 links that users of Meta's Muse AI agent had created to share conversations, and that opening a few hundred turned up material clearly not meant to be public, Ars Technica reports. Examples include a candidate's notes for a confidential call with a frontier AI lab, a company's internal revenue review, a small-business loan application and a staff vacation calendar.

The cause, Ars says, is that Muse puts the conversation title in the link's subdomain on a single Meta domain. When a recipient opens the link, the name passes through DNS and can end up in commercial passive-DNS databases that anyone can search. Claroc calls these "GhostShares." "It took one query," the researchers wrote.

Claroc's CTO says he reported the issue through Meta's bug bounty on October 8. Meta told Ars it has paused the sharing feature while it reviews the report and considers changes to how links can be discovered.

Attackers compromised third-party operators of the .GH (Ghana), .SL (Sierra Leone) and .AS (American Samoa) country-code domains, changed DNS records and obtained valid HTTPS certificates for domains they didn't own, including several Google domains, Google says. Google says its own systems were not compromised. It blocked the certificates in Chrome, worked to have them revoked, and found other organizations, including major brands, affected by the same attacks. Chrome users don't need to do anything; Google notes other browsers may not be protected.