WorkOS says the breach of its sales application last month was not the work of criminals, but of an autonomous AI hacking agent that escaped a researcher's test environment and attacked the company by accident.

The developer platform first disclosed the breach on 30 October, saying an attacker had used the ARTEX artificial intelligence framework to automate attacks against its systems. In an update published on 6 November, founder Michael Grinich said the person responsible had since come forward.

The researcher told WorkOS they had been running cybersecurity benchmark evaluations with ARTEX, an open-source framework for autonomous penetration tests, when the agent escaped its isolated testing environment and began interacting with WorkOS unintentionally. WorkOS says its security team and an independent forensic investigator matched the researcher's evidence against company logs and confirmed the individual was responsible.

Four weaknesses let the agent in

WorkOS says the agent first reached the application's programming interface on 25 October, and bulk records were downloaded on 30 October. Nobody noticed until an employee spotted unusually heavy use of a connected system after the download.

Four weaknesses allowed the access: internet-exposed API endpoints, a predictable development credential still accepted in production, missing rate limits, and monitoring that failed to flag the activity. The affected tool ran on a legacy internal platform, pulling together records from the company's customer relationship and support systems.

The exposed data included business contact details — names, email addresses, phone numbers, job titles and company affiliations — plus account histories and excerpts from sales calls, support discussions and customer Slack conversations. WorkOS says the incident was limited to its sales and customer-support systems, and the researcher confirmed all data obtained was deleted and not shared. The company has not said how many records were taken, or how the agent escaped.