WorkOS says an escaped AI testing agent reached its internal sales app and exposed customer records
The breach went unnoticed for five days, and the company is now testing its own systems with the same framework that was involved.
WorkOS has said an AI-driven intrusion into its internal sales application, which exposed customer records, was carried out by a security researcher who now says an autonomous testing agent escaped its sandbox and reached WorkOS by accident. The company's forensic review matched the researcher's artifacts against its logs, and the researcher reports that all data taken has been deleted. The investigation found four weaknesses: a legacy internal tools platform without modern protections, API endpoints exposed to the internet, a predictable development credential left live in production, and no rate limiting. The first unauthorized access happened on October 25 but went unnoticed until October 30, when an employee spotted unusual usage of a connected system. WorkOS has not yet explained how the agent got out of its test environment, and it has since begun testing its own systems with the same framework.
Why it matters
The incident shows that a security testing tool can reach live customer data when test boundaries fail, and that a single forgotten development credential in production was enough for access. Customers whose business contact details, account histories and support communication excerpts were exposed now face a question about how their information was handled and who else may have seen it. The researcher's report that all data was deleted and not shared has been matched against WorkOS logs, but how the agent escaped its sandbox remains unexplained.
Matters or hype?
Does this story matter, or is it hype? Vote before you see what everyone else said.