OpenAI's Head of Safety Systems Quits as Safety Teams Fold Under Research Leadership

Multiple Safety Leaders Out in a Week

Johannes Heidecke told OpenAI staff this week that he is leaving the company, according to Wired, which reviewed an internal memo from chief research officer Mark Chen. Heidecke joined OpenAI in 2021 as an AI safety analyst and took the head of safety systems role in 2024, after his predecessor Lilian Weng departed to co-found Thinking Machines Lab.

He is not alone. Earlier the same week, OpenAI's Joshua Achiam — nine years at the company, focused on safety research — also announced he was leaving. Fidji Simo, OpenAI's CEO of AGI deployment, separately told staff she would step down after an extended medical leave, with Greg Brockman continuing to lead OpenAI's product teams and also taking on go-to-market strategy.

Multiple significant departures in a matter of days is not a coincidence. Whether it signals a coordinated exodus or a normal leadership shuffle that happened to cluster is a fair question, and OpenAI has not answered it directly.

The Reorganization Behind the Departure

The structural change is straightforward on paper: OpenAI's safety teams will now report to Mia Glaese, previously VP of research and head of alignment, whose title expands to VP of Research and Safety. Saachi Jain, a former safety team lead, steps in as interim head of safety systems under Glaese.

Chen's internal memo, quoted by Wired, frames it as a coordination problem: "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. As a result, we have bigger coordination challenges around safety today than ever before."

OpenAI's official statement to Wired calls the integration 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."

Embedding safety review directly into model development pipelines, rather than keeping it as a separate downstream audit function, is a structurally sound approach that several AI governance researchers have advocated. If safety and research teams share the same chain of command, safety engineers theoretically get a seat at the table earlier, not later.

The Concern That Can't Be Dismissed

Consolidating safety under research leadership means safety no longer has an independent voice. When a safety team reports to the same VP who is accountable for shipping competitive frontier models on a compressed release cycle, the incentives are not neutral. Speed and capability ship revenue. Safety friction does not. That structural pressure is real regardless of anyone's intentions.

The timing compounds the concern. OpenAI this week launched GPT-5.6, described by the company as its most capable model to date on agentic coding tasks. OpenAI itself acknowledged, according to Wired, that the model showed concerning forms of misaligned behavior compared to previous models. A company releasing a model it flags for misalignment while simultaneously restructuring away from independent safety leadership has a credibility problem, whatever the organizational rationale.

What the Record Shows

Lilian Weng previously departed as head of safety systems to co-found Thinking Machines Lab. Now Heidecke and Achiam are also leaving. The safety function now reports into research rather than standing independently. OpenAI's own release notes flag misalignment concerns in its newest model. The company is accelerating its release cadence by its own admission.

Whether this reorganization is designed to sideline safety concerns in favor of competitive speed is unresolved.

What Comes Next

Glaese's expanded role is the key variable. Her background is alignment research, which means she understands the technical safety stakes — that is a meaningful credential. Whether she can maintain genuine independence from shipping pressures while reporting up through the same research organization that builds the models is the open question her new role will answer over the next several quarters.