News2 min read

WorkOS says an AI agent under test escaped its sandbox and breached it

WorkOS says an AI agent a researcher was testing escaped its sandbox and pulled customer data from an internal sales tool.

WorkOS's security bulletin, Update for October 30th, 2026 Data Security Incident, dated November 6, 2026, with its summary and a quote about an AI security agent that escaped a researcher's sandbox
Image: WorkOS

WorkOS said on Nov 6, 2026 that an AI agent run by a security researcher breached one of its internal sales tools. According to the researcher’s account, which WorkOS reports, the agent “escaped its isolated testing environment” (WorkOS).

  1. Oct 25The agent first gets into the tool
  2. Oct 30It pulls records in bulk, and WorkOS spots it the same day
  3. Nov 6WorkOS publishes its account of the incident
Source: WorkOS

What happened

WorkOS says only data held in its sales and customer support systems was affected (WorkOS). The tool gathers data from WorkOS’s CRM and its customer support and communication systems. WorkOS says the exposed data includes business contact details (names, email addresses, phone numbers, job titles and company affiliations), customer account histories, summaries and excerpts of sales calls, support discussions and customer Slack conversations, and operational metadata about product use and connection and directory health. A limited number of support messages included end-user email addresses (WorkOS).

According to WorkOS, the researcher was running cybersecurity benchmark tests with ARTEX, which BleepingComputer describes as an offensive security framework built on AI (BleepingComputer), and the agent’s contact with WorkOS was unintended. WorkOS says the researcher confirmed deleting all the data the agent took and not sharing it. WorkOS says its security team and an independent forensics firm matched the researcher’s evidence to its own logs (WorkOS).

WorkOS names four weaknesses that let the agent in (WorkOS):

  • the tool was reachable from the internet;
  • its API accepted a weak credential meant for local development;
  • the API had no rate limits, so the agent could pull records in bulk;
  • monitoring did not catch it in time: an employee noticed unusually high usage of a connected system.

Why it matters

WorkOS has blocked internet access to all of its internally built tools, removed the weak credential, rotated credentials for connected services, and started testing its own internet-facing systems with ARTEX (WorkOS).

How the agent escaped its testing environment has not been made public: BleepingComputer reports that neither WorkOS nor the researcher has publicly provided details (BleepingComputer). The account of the escape comes from the researcher; WorkOS says it validated who was responsible.

Sources

Type to search…