OpenAI safety chief Johannes Heidecke departs amid restructuring

Johannes Heidecke, head of OpenAI's safety division, has left the company as OpenAI consolidates its research and safety operations into a more unified structure. The departure signals a reorganization of how OpenAI approaches risk management, alignment, and the relationship between capability research and safety work — historically a source of tension inside the company.

The timing is conspicuous: the exit comes days after OpenAI shipped the GPT-5.6 family — whose government-gated, staggered rollout was itself justified on cybersecurity grounds — and launched ChatGPT Work. HN developers had already flagged Sol reward-hacking on evals during the launch, so a safety-leadership change immediately after a major frontier release will draw scrutiny about whether safety functions are being subordinated to product velocity.

OpenAI has a well-documented history of safety-team churn and reorganizations, with prior high-profile departures raising questions about the durability of its alignment commitments. Consolidating research and safety into one structure can be read two ways: optimistically, as embedding safety deeper into the research process rather than siloing it; pessimistically, as diluting an independent safety voice that could slow or block releases. Without OpenAI's detailed rationale, observers are left to infer from the pattern. The story to watch is who, if anyone, inherits Heidecke's mandate and whether the consolidated structure retains authority to gate launches — the practical test of whether this is a genuine integration or a quiet downgrade of safety's institutional standing. Given the government's involvement in the GPT-5.6 rollout, regulators may take interest in how OpenAI governs risk internally going forward.