OpenAI's head of safety, Johannes Heidecke, is leaving as OpenAI integrates its research and safety teams; Mia Glaese will become VP of research and safety
Johannes Heidecke's departure comes as OpenAI tries to further integrate its research and safety teams.
Context & Ripple Effects
OpenAI is consolidating research and safety under a single VP role as Johannes Heidecke departs and Mia Glaese takes responsibility for the combined organization. The change follows other safety- and alignment-related exits in the coverage, including Andrea Vallone’s planned departure and the earlier disbanding of the mission alignment team led by Joshua Achiam.
The arc is less a standalone personnel move than a reorganization of how OpenAI locates safety work within its research operation. That makes the successor’s remit—and the degree of independence retained by safety functions—material to how the company governs model development.
First-order effects
- Mia Glaese assumes a broader VP-level mandate spanning both research and safety, while Heidecke’s departure removes the current head of safety from the organization.
- OpenAI’s safety work is moved into a more integrated operating structure with research rather than remaining a separately led function.
Second-order effects
- Teams working on model development and safety evaluation will have to align priorities, reporting lines, and release processes under one leader; this can reduce organizational handoffs but concentrates trade-off decisions.
- The sequence of departures from safety and alignment roles increases scrutiny of whether OpenAI can retain specialized safety leadership during organizational change.
Third-order effects
- If major AI developers continue to fold safety organizations into core research, AI governance may increasingly depend on internal product-and-research decision processes rather than visibly separate safety chains of command.
- The durability of this model will hinge on whether integrated teams preserve the authority and resources to challenge deployment decisions; the available coverage does not establish how OpenAI will structure those safeguards.
The trend: AI labs are increasingly treating safety as a function embedded in frontier-model research, even as leadership turnover tests the institutional independence of that work.