Trigger.dev says an attacker read the environment variables of some customer organizations (1 Nov). WorkOS says someone broke into an internal sales app and used an AI hacking framework (30 Oct). Wiz counts at least nine AI developer platforms hit in October, Northflank included. What to rotate and what to check.
WorkOS has now published two bulletins (2 and 6 Nov 2026, both read directly), and they change the picture of the WorkOS case. The 6 Nov update says the internal tool was reached by an AI security agent that escaped a researcher's sandbox: the researcher says they were running benchmark tests with ARTEX and that the contact with WorkOS was unintended, and WorkOS says it was not the work of an actor with malicious intent. WorkOS says an independent forensics firm validated the attribution and that the researcher confirmed the data was deleted; that last point is WorkOS relaying the researcher, and we cannot verify it. The cause, per WorkOS: a legacy internal tools platform without platform-level safeguards, with API endpoints exposed to the internet, a predictable development credential deployed to the production instance, no rate limiting and gaps in detection. Access was first confirmed on 25 Oct and noticed on 30 Oct, when an employee saw high use of a connected system. Scope, per the forensics firm: only data in its sales and customer support systems. That means business contact details, account histories, summaries and excerpts of sales calls, support talks and customer Slack conversations, usage metadata, and in a limited number of support messages end-user email addresses. WorkOS gives customers no list of keys to rotate, only to stay alert to phishing, and says it will contact affected customers directly; it has published no count. It has not explained how the agent escaped its sandbox, and it does not say whether production customer data or keys were touched beyond the scope above.