Skip to content
NewsLeadership

OpenAI Folds Safety Team into Research Again as Head Johannes Heidecke Exits

· by Pondero Newsdesk

The short version

Johannes Heidecke is leaving OpenAI after two years as head of safety systems following a reorganization that embeds safety under research leadership for the second time in under two years.

OpenAI Folds Safety Team into Research Again as Head Johannes Heidecke Exits

For the second time in under two years, OpenAI has absorbed its safety organization into a structure reporting to research leadership. Johannes Heidecke announced his departure as head of safety systems on July 11, 2026, the same day Chief Research Officer Mark Chen told staff the restructuring would take effect.

What happened

Mark Chen's memo, seen by Wired, said safety teams will now report to Mia Glaese, whose title expands to VP of Research and Safety under the new structure. Saachi Jain takes over as interim head of safety systems while OpenAI searches for Heidecke's permanent replacement.

Chen's stated rationale, per Engadget's report, was embedding safety inside research to give it "an earlier and more direct role in shaping key model, product and launch decisions." Heidecke joined OpenAI in 2021 as an AI safety analyst and took the safety systems role in 2024, succeeding Lilian Weng.

Context and prior departures

The Next Web confirmed this is the second time in less than two years OpenAI has placed its safety function inside the research org. The first instance was May 2024, when the Superalignment team dissolved after co-leads Ilya Sutskever and Jan Leike departed. Leike wrote publicly on leaving that "safety culture and processes have taken a backseat to shiny products."

The AGI Readiness team leader resigned in October 2024. The Mission Alignment team, launched as Superalignment's successor, was disbanded in February 2026 after 16 months. April 2026 brought the departure of OpenAI's product chief, the head of Sora, and its enterprise CTO on a single day. Fidji Simo, chief of applications, stepped down this month citing a prolonged medical recovery.

Lilian Weng, who held the safety systems role before Heidecke, now works at Thinking Machines Labs, the startup founded by former OpenAI CTO Mira Murati.

The restructuring arrives as 42 state attorneys general have opened an investigation into OpenAI, serving a subpoena on advertising, user data, and internal policies shortly after the company filed confidentially for a stock market listing, per TNW.

Why it matters

Safety at OpenAI now sits inside the same organizational hierarchy that owns model shipping decisions. A safety team embedded in research reports through the same chain as the teams building and releasing the models it evaluates. Critics of this structure, including Jan Leike in his 2024 departure statement, argued it reduces the independence safety reviewers need to delay or block a product on safety grounds.

For teams that factor OpenAI's published safety evaluations into procurement or compliance decisions, the org change is relevant: the evaluation function now operates within the research organization rather than separately. OpenAI's CEO Sam Altman has noted a separate Head of Preparedness position focused on severe AI-related risks, but that role's structural independence under the new reporting lines is not yet documented publicly.

What to watch next

Heidecke's permanent replacement hire and that person's reporting line will signal how much structural independence OpenAI preserves in safety systems. The S-1 prospectus, once public, will formally describe the post-restructuring governance structure to investors. The first model release following the reorganization is the practical test of whether embedding safety inside research changes evaluation timing or methodology.

Sources