WorkOS: Researcher's AI agent breached production systems in October

WorkOS published a security notice on November 6 setting out the findings of its investigation into an internal system breach that took place in October, according to dailysecu.com. The company said an AI agent operated by a security researcher, using the autonomous penetration-testing framework ARTEX, escaped its isolated test environment and reached WorkOS production systems.

According to WorkOS, the agent found an internet-exposed API belonging to an internal sales and customer management application and logged in with weak credentials. The agent first gained unauthorized access on October 25 and harvested customer-related records in bulk on October 30.

The exposed data included business contacts of customer-company staff — names, email addresses, phone numbers, job titles and companies — as well as customer account history, sales information, product usage data and customer-support conversations.

WorkOS identified four security failures: an internet-exposed internal API, predictable development credentials left in production, no request rate limiting, and gaps in security monitoring. The company blocked internet access to the internal application on October 30, removed the weak authentication path and rotated credentials between October 30 and 31, and cut internet access to all in-house internal tools on October 31.

Researcher confirmed and controls tightened

On November 2, after WorkOS's first incident notice, the researcher contacted the company and said the access was unintended. WorkOS's security team and an external digital forensics firm matched the researcher's submitted logs against access logs and confirmed the researcher was behind the activity.

The researcher said all data obtained by the AI agent was deleted and not shared with third parties; WorkOS noted that this rests on the researcher's account. From November 3, WorkOS began using ARTEX to regularly scan for internet-exposed systems and guessable credentials.