OpenAI Head of Safety Leaves Amid Reorganization
OpenAI’s head of safety systems, Johannes Heidecke, announced his departure to staff this week, according to internal communications obtained by WIRED. Heidecke’s exit comes as the company implements a restructuring that aims to more closely integrate its safety and research teams.
Reorganization Details
In a memo to staff seen by WIRED, chief research officer [name not provided in source] stated that OpenAI’s safety teams will now report to the company’s VP of research and head of alignment, Mia Glaese, who is taking on an expanded role as VP of research and safety. Saachi Jain, who previously led safety teams at OpenAI’s No. 2 Executive, 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,” said Chen in the memo. “As a result, we have bigger coordination challenges around safety today than ever before.”
Integration of Safety and Research
The reorganization reflects OpenAI’s push to embed safety work more directly into frontier-model development. “It’s 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,” Chen stated. “We’re excited for this next chapter under Mia Glaese’s leadership across research and safety.”
Departures and Changes
Heidecke joined OpenAI in 2021 as an AI safety analyst and became head of safety systems in 2024, succeeding Lilian Weng, who left to cofound Thinking Machines Lab. His departure follows a series of leadership changes at the company.
Other Notable Departures
- OpenAI’s chief futurist, Joshua Achiam, also told colleagues he would leave after nine years researching safety, earlier this week.
- CEO of AGI deployment Fidji Simo informed staff she would step down after an extended medical leave. She will remain as a part-time adviser. Greg Brockman will continue leading her former responsibilities and also take on go-to-market strategy.
These moves coincide with the launch of GPT-5.6, OpenAI’s most capable model to date on agentic coding tasks. However, compared to previous models, OpenAI says GPT-5.6 showed .
The article was reported by Maxwell Zeff, a senior writer at WIRED covering the business of artificial intelligence.