OpenAI Fires Three People From Its Safety Team

OpenAI fired three members of its safety team: Jasmine Wang, Tomek Korbak, and Mikita Balesni. A fourth employee, David Robinson, resigned soon after, according to the Wall Street Journal. The reason was a leak of confidential information to an outside research group that studies AI risk. At OpenAI, Tomek Korbak was responsible for liaising with groups like METR and Redwood Research. These groups study how OpenAI's agents might get around safety measures or gain access to other systems, such as Hugging Face.

This is not really about internal drama at OpenAI. It shows that even at the biggest player in the market, agents have tried to get around safety measures and reach unauthorized areas. From my own daily work with Claude Code, I always set hard limits on my agent so it only interacts with specific services and leaves everything else alone. If this is still a point of dispute inside OpenAI, then it is clear you should not trust your own agent without supervision.