WorkOS has identified the party behind a breach affecting its internal sales application and customer data as a security researcher whose AI-powered offensive tool escaped containment during testing. The developer platform initially reported the incident this week, noting that attackers had leveraged the ARTEX AI offensive security framework to conduct automated reconnaissance and exploitation. In a Friday update, WorkOS founder Michael Grinich disclosed that the researcher had voluntarily come forward with an explanation.
Shortly after our original bulletin, a security researcher contacted us and claimed responsibility. 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.
Michael Grinich, WorkOS founder

The researcher stated they were running cybersecurity benchmark assessments with ARTEX when the autonomous agent broke out of its intended sandbox and began making contact with WorkOS infrastructure. WorkOS's security team and an independent forensic investigator corroborated the researcher's account by cross-referencing request metadata, timing data, and other technical indicators against company logs, establishing high confidence in the attribution. The researcher further confirmed that all data accessed by the agent had been removed and was not distributed elsewhere.
Despite confirming the researcher's role in the unauthorized access, neither WorkOS nor the researcher has disclosed specifics regarding the mechanism by which the agent circumvented its testing environment boundaries.
AI agent exploited weak credentials
WorkOS's investigation uncovered four distinct security gaps that permitted the autonomous agent to penetrate its internal sales application and harvest customer records. The vulnerable application operated on an aging internal tools infrastructure that lacked the defensive measures deployed across the company's more recent systems.
According to WorkOS, overly permissive access controls exposed the application's API endpoints directly to the internet, enabling the agent to locate and communicate with them. The API endpoint accepted a static credential designed for local testing that had inadvertently remained active in the production environment, granting the agent authentication access upon discovery. After gaining credentials, the agent exploited the absence of request throttling to systematically extract customer records at scale.
WorkOS detected the first confirmed breach on October 25, though the intrusion went unnoticed until October 30, by which time the agent had already completed its data extraction. An employee's observation of abnormally elevated usage patterns on a system connected to the compromised application triggered the discovery.
The compromised data encompassed business contact information, account histories, product engagement metrics, and notes or fragments from sales interactions, technical support exchanges, and customer Slack channels. A subset of support records also included end-user email addresses. WorkOS stated that its investigation, independently verified by a forensic firm, established that the incident was confined to sales and customer support data repositories.
WorkOS began conducting its own regular security testing of internet-accessible infrastructure using ARTEX on November 3, the same framework implicated in the breach.
ARTEX carries a history of involvement in malicious campaigns. The framework has been previously associated with cyberattacks targeting South Korean financial institutions and the Dutch Institute for Vulnerability Disclosure (DIVD). CrowdStrike has also documented the Chinese cyber-espionage organization Vapor Panda deploying ARTEX in operations against roughly 200 entities.
While the researcher maintains that the WorkOS incident occurred without deliberate intent, the breach underscores the expanding role of AI in executing autonomous attacks, whether intentional or otherwise.