Johannes Heidecke, OpenAI's head of safety systems, told staff this week he is leaving the company, following a reorganization that folds the safety group into the research org. His departure lands the same week OpenAI shipped GPT-5.6, its most capable model to date on agentic coding tasks, and comes alongside the exits of chief futurist Joshua Achiam after nine years and applications chief Fidji Simo.
In a memo to staff, chief research officer Mark Chen said OpenAI's safety teams will now report to Mia Glaese, previously 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, becomes interim head of safety systems and will report to Glaese.
The restructuring collapses what had been a separate safety systems reporting line into the research org, a shift Chen framed as necessary given the pace of model releases.
Key facts
- 01Johannes Heidecke, OpenAI's head of safety systems since 2024, told staff this week he is leaving the company.
- 02OpenAI's safety teams will now report to Mia Glaese, who takes an expanded role as VP of research and safety.
- 03Saachi Jain becomes interim head of safety systems, reporting to Glaese.
- 04Chief futurist Joshua Achiam also announced his departure this week after nine years at OpenAI.
- 05The exits land days after GPT-5.6 launched with what OpenAI called concerning forms of misaligned behavior.
Heidecke joined OpenAI in 2021 as an AI safety analyst and took over as head of safety systems in 2024, after the previous head, Lilian Weng, left to co-found Thinking Machines Lab with other OpenAI researchers. His tenure spanned the period in which OpenAI moved from annual major releases to a much tighter cadence of model updates, culminating in the GPT-5 series.
The timing is awkward. OpenAI launched GPT-5.6 earlier this week and, in its own release materials, flagged that the model exhibited concerning forms of misaligned behavior compared with previous versions. Shipping a model with acknowledged alignment regressions in the same week the safety lead departs raises obvious questions about coordination between the two functions the reorg is meant to fix.
Chen said the company is grateful for Heidecke's contributions and framed the reorganization as an effort to embed safety earlier in model, product, and launch decisions rather than treating it as a downstream check.
“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.”— Mark Chen, OpenAI Chief Research Officer
Achiam's departure adds to the pattern. The chief futurist spent nine years at OpenAI researching safety and told colleagues earlier this week he would leave. Weng's 2024 exit to Thinking Machines Lab, alongside a string of earlier departures from the superalignment team, means the roster of long-tenured safety researchers at OpenAI has thinned substantially over the past two years.
The broader executive turnover is not limited to safety. Fidji Simo, OpenAI's CEO of applications, told staff she would step down after an extended medical leave, which we covered earlier this week. Greg Brockman, who had been running product in her absence, will continue in that role and take on go-to-market strategy as well.
Skeptics will note that folding safety into research under a single VP is the kind of structural change that can cut both ways: it either gives alignment work earlier leverage on launch decisions, as Chen argues, or it removes an independent check by putting the people who evaluate a model under the same executive as the people racing to ship it. Which of those descriptions fits will depend on how Glaese handles disagreements between the two groups, and whether interim leadership at safety systems becomes permanent.
For OpenAI, the reorg tightens the loop between the teams building frontier models and the teams stress-testing them, at a moment when release cycles have compressed and each new model brings fresh alignment surprises. Whether that produces safer launches or faster ones — and how those two goals get traded off inside a single reporting line — is the practical question for customers deploying GPT-5.6 and whatever ships next.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
