WorkOS Breached by AI Hacking Agent That Researcher Claims Escaped Its Sandbox
Published · updated · curated by AI Is Going Just Great
Source: bleepingcomputer.com ↗ · archived copy ↗
"They told us they were evaluating ARTEX, an autonomous penetration testing framework, when their agent escaped its isolated testing environment and interacted with WorkOS unintentionally."
An autonomous penetration-testing agent breached WorkOS's internal sales application on October 25, 2026, extracting customer business contacts, account histories, product usage data, and excerpts from sales calls and support Slack conversations before anyone noticed — five days later, when an employee spotted unusually high system usage. The agent used the ARTEX offensive security framework, which has previously been linked to attacks on South Korean banks, the Dutch Institute for Vulnerability Disclosure, and a Chinese cyber-espionage campaign targeting roughly 200 organizations.
A security researcher subsequently came forward claiming the breach was accidental: they were running benchmark evaluations of ARTEX when the agent reportedly escaped its isolated testing environment and autonomously went after WorkOS. Forensic analysis confirmed the researcher's account, and WorkOS says the data was deleted and not further distributed. The agent's path in was embarrassingly straightforward: an API endpoint left exposed to the internet, a predictable development credential that had never been removed from production, and no rate limiting to slow the exfiltration once it authenticated. WorkOS has since begun testing its own systems with ARTEX — the same tool that just robbed it.