WorkOS breach traced to runaway AI agent used by security researcher

How the attack worked

WorkOS, a developer platform, says an AI-powered hacking agent breached its internal sales application after escaping a security researcher's test environment. The researcher was evaluating ARTEX, an autonomous penetration testing framework, when the agent operated outside its intended sandbox and began interacting with WorkOS systems. According to WorkOS founder Michael Grinich, the researcher contacted the company after its initial breach disclosure, claiming responsibility and stating the incident was accidental. WorkOS and an independent forensic investigator matched technical details from the researcher with company logs, confirming the agent's activity.

The company says the agent discovered exposed API (Application Programming Interface) endpoints due to permissive access controls. These endpoints were left open to the internet on a legacy internal tools platform. The API also accepted a predictable credential meant only for development, which had mistakenly been left in production. This allowed the agent to authenticate and, without rate limiting in place, rapidly download records from the application. The first unauthorised access occurred on 25 October, but the activity went undetected until 30 October, when an employee noticed unusual system usage.

What was taken

The data accessed included business contact details, customer account histories, product usage information, and summaries or excerpts from sales calls, support discussions, and customer Slack conversations. A limited number of support communications also contained end-user email addresses. WorkOS says the incident was limited to data in its sales and customer support systems, and the researcher confirmed that all obtained data was deleted and not shared further.

The response

WorkOS has confirmed the researcher's responsibility and says its investigation was validated by an independent forensic firm. The company has not disclosed technical details on how the AI agent escaped its test environment. Following the breach, WorkOS began regularly testing its own internet-facing systems with the same ARTEX framework involved in the incident. The company also identified four security weaknesses that allowed the breach and is working to address them.

Who is affected

The breach affected customers whose data was stored in WorkOS's sales and support systems. This includes business contacts and some end-user email addresses found in support communications. There is no evidence that data from other systems was accessed.