Meta paused Muse share links after researchers found about 6,800 of them in public DNS. Check how your automations share files

A share link is a publishing decision, and Meta's Muse just showed how quickly a private link stops being private.

What happened

Security firm Claroc says it enumerated about 6,800 links that Muse users created to share conversations. Muse built each share address with the conversation's title in the subdomain, for example a pattern like q3-revenue-review plus a random string on cf.metaaiusercontent.com. Once someone opened a link, the hostname landed in passive DNS databases, which anyone can search through commercial services. The researchers told Ars Technica it "took one query."

After opening a few hundred, they found a candidate's notes for a confidential executive search call, a marketing company's internal revenue review, a small business loan application, a real estate investor's return and mortgage figures, and a full year staff vacation calendar. Claroc says it reported the issue through Meta's bug bounty on October 8. Meta told Ars and The Verge it has paused the sharing feature while it reviews the report.

My take

The bug is specific to Muse, but the pattern is everywhere in small business automation. Workflows generate a report, drop it in a folder, and send an "anyone with the link" URL to a client or a Slack channel. The file name often says exactly what is inside: client name, invoice, payroll, Q3 numbers.

When I build these flows I treat every generated link as public by default and design from there:

1. Keep sensitive words out of file names, folder names and URLs. Use IDs, and put the descriptive title inside the document.
2. Share to named accounts where the tool allows it, not open links.
3. Set expiry on links that must be open, and have the workflow revoke them after the job is done.
4. Never paste raw financials or personal data into an AI chat you plan to share. Send a summary.
5. Log every link the automation creates, so you can audit and kill them in one pass.

None of this is hard. It just has to be decided before the workflow ships, not after a researcher finds it.