Agent without guardrails is not rogue. Rogue is something else. In this case OpenAI deliberately launched an agent to do action A, but it went further and breached the website. Feels more like a car accident which hit government building.

Not yet peer reviewed