OpenAI's safety chief is out, its newest model is showing troubling behavior, and the company is reorganizing safety to be more deeply embedded in model development. Those three things happened in the same week.
Johannes Heidecke, OpenAI's head of safety systems, told staff this week he is leaving the company, WIRED reported.
His departure is part of a broader reorganization. Chief research officer Mark Chen told staff in a memo that safety teams will now report to Mia Glaese, the company's VP of research and head of alignment, who takes on an expanded role as VP of research and safety.
Saachi Jain, who previously led safety teams at OpenAI, will serve as interim head of safety systems reporting to Glaese.
"The demands on safety continue to increase. We are training models at a much faster cadence, and release cycles have come down greatly in turn," Chen wrote in the memo. "As a result.
We have bigger coordination challenges around safety today than ever before." The reorganization comes days after OpenAI launched GPT-5.6, its most capable model for agentic coding tasks. The company acknowledged the model showed "concerning forms of misaligned behavior" compared to previous versions, per WIRED.
OpenAI restricted access to a selective group of strategic partners, and the launch was delayed at the request of the US government over security concerns.
Heidecke is not the only safety-focused leader heading for the exit. OpenAI's chief futurist, Joshua Achiam, told colleagues earlier this week he would leave after nine years researching safety. And Fidji Simo, the company's CEO of AGI deployment, said she would step down after an extended medical leave, with Greg Brockman taking over go-to-market strategy alongside his product leadership role.
Chen said in a statement that it is "important that our safety work is integrated with frontier-model development, with an earlier and more direct role in shaping key model, product, and launch decisions." The safety shakeup is unfolding as OpenAI also escalated its bio-defense efforts. The company doubled its maximum bug bounty award to $50,000 for researchers who can develop a universal jailbreak bypassing GPT-5.6's biological safeguards, part of an ongoing private Bio Bounty program running through July 27.
