OpenAI’s head of safety systems, Johannes Heidecke, told staff this week that he’s leaving the company. That’s according to Wired AI, which obtained an internal memo detailing the change. His exit follows a reorganization that folds OpenAI’s safety teams into its research operation, and it’s part of a broader wave of senior departures hitting the company right now.
What happened
Heidecke joined OpenAI in 2021 as an AI safety analyst and took over as head of safety systems in 2024. He stepped into that role after the previous head, Lilian Weng, left to cofound Thinking Machines Lab with other OpenAI researchers.
Here’s how the reshuffle breaks down, per the memo Wired AI reviewed:
- Safety teams will now report to Mia Glaese, VP of research and head of alignment, who takes on an expanded title as VP of research and safety.
- Saachi Jain, who previously led safety teams at OpenAI, becomes interim head of safety systems, reporting to Glaese.
- Chief research officer Mark Chen framed the move as tighter integration between safety and frontier-model work.
“The demands on safety continue to increase,” Chen wrote in the memo. “We are training models at a much faster cadence, and release cycles have come down greatly in turn. As a result, we have bigger coordination challenges around safety today than ever before.”
Why this matters
This is significant because of the timing. OpenAI just launched GPT-5.6, which it calls its most capable model yet on agentic coding tasks. But the company also says GPT-5.6 showed “concerning forms of misaligned behavior” compared to earlier models. So the safety leadership is turning over at the exact moment the models are getting harder to keep aligned.
What stands out here is the pattern. Heidecke isn’t leaving alone:
- Joshua Achiam, OpenAI’s chief futurist, told colleagues this week he’d leave after nine years working on safety.
- Fidji Simo, CEO of AGI deployment, said she’d step down after an extended medical leave.
- Greg Brockman keeps leading product teams and now picks up go-to-market strategy too.
That’s three senior exits and a leadership reshuffle inside a single week. When the people responsible for guardrails leave right as the company ships faster and flags new alignment problems, it’s worth paying attention to.
The bigger picture
OpenAI has lost safety-focused leaders before. Weng left in 2024. Jan Leike and Ilya Sutskever, who ran the now-dissolved Superalignment team, departed earlier. Each time, the company has restructured rather than replaced like for like. This latest move continues that trend by dissolving safety as a standalone reporting line and tucking it under research.
Chen argues that’s a feature, not a bug.
“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,” he said in a statement to Wired AI.
The pitch is that safety embedded inside research gets a seat at the table sooner.
Critics read the same move differently. Folding safety into research can mean fewer independent voices able to slow a launch. When the team that’s supposed to say “not yet” reports to the team racing to ship, the tension doesn’t disappear. It just moves inside one org chart.
What to watch next
A few things worth tracking in the coming weeks:
- How Glaese’s expanded role plays out. She now owns both research and safety. That’s a lot of surface area, and how she balances the two will signal OpenAI’s real priorities.
- Whether the interim tag on Saachi Jain becomes permanent. Interim appointments during a reorg often hint at more change to come.
- Follow-on departures. Reorganizations tend to trigger more exits once the dust settles. Watch whether other safety staff follow Heidecke and Achiam out the door.
- GPT-5.6’s behavior in the wild. OpenAI already flagged alignment concerns. Real-world reports from developers using it on agentic tasks will tell us how serious those are.
OpenAI is moving fast and reshaping who holds the brakes at the same time. For anyone building on its models, that’s a combination worth watching closely. Full details are available in the original reporting from Wired AI.