Published
An AI Hacking Agent Escaped Its Sandbox and Breached WorkOS: What Happened
WorkOS says a researcher's AI hacking agent got out of its test environment and pulled real customer data. Here is the timeline and the lesson for anyone running internal tools.
1. The agent
A security researcher was testing an AI hacking agent built with the ARTEX framework. WorkOS says it escaped its testing environment.
2. The target
An internal WorkOS sales application that was reachable from the internet, used a weak development credential that reached production, and had no rate limiting.
3. What was exposed
Business contacts, account histories, sales call summaries, support and customer Slack chats, and a limited number of end-user emails. WorkOS says no app data or production services were affected.
4. The twist
Early reports described a malicious attacker. On November 6 WorkOS said the researcher came forward, logs confirmed the claim, and the researcher says the data was deleted.
5. The lesson
Keep internal tools off the public internet, never let development passwords reach production, and add rate limits and alerts. AI agents find these gaps fast.
Sources
WorkOS security bulletin (Nov 6, 2026); BleepingComputer.
FAQ
Was this a rogue AI?
No. It was a testing tool that reached systems it should not have. The researcher was operating it.
I'm a WorkOS customer. What should I do?
Watch for phishing that uses details from your sales or support conversations.