OpenAI's head of safety is walking out the door.
Johannes Heidecke is departing OpenAI at a moment the company describes as an effort to more tightly integrate its research and safety functions. The structural change — folding safety work closer to the teams building the models — is framed internally as a collaboration move. Whether Heidecke's exit is a cause of that shift, a consequence of it, or simply coincidental timing, OpenAI has not said.
The departure matters because safety leadership at AI labs is increasingly symbolic as well as operational. When a head of safety leaves, it tends to generate pointed questions about whether safety concerns are being subordinated to shipping velocity — questions OpenAI has faced before, notably after several high-profile resignations in 2024. Integrating safety and research teams can mean closer collaboration, but it can also mean safety loses independent standing to push back.
OpenAI has repeatedly insisted that safety and commercial progress are compatible goals. The pattern of safety-team exits suggests at least some insiders disagree.