OpenAI's head of safety, Johannes Heidecke, is leaving the company as part of a broader reorganization that merges the company's safety and research teams under a single executive. The move signals a shift in how the artificial intelligence firm manages safety oversight.
Safety and Research Teams to Merge Under One Leader
According to a report from WIRED, the reorganization places OpenAI's safety teams under the leadership of Mia Glaese, who currently serves as vice-president of research and head of alignment. Glaese's role has been expanded to oversee both research and safety, effectively combining two previously separate functions.
The report, based on a company memo seen by WIRED, states that chief research officer Mark Chen announced the new structure. Saachi Jain will step in as interim head of safety systems and will report directly to Glaese.
Heidecke's Departure Follows Internal Changes
Heidecke informed colleagues this week that he was leaving the company, according to The Edge Malaysia, which cited the WIRED report. His departure comes as OpenAI continues to evolve its internal structure amid growing scrutiny over AI safety practices.
The reorganization is designed to streamline decision-making by having one executive responsible for both advancing AI capabilities and ensuring those systems are safe. This integration of research and safety teams is a notable change for OpenAI, which has historically kept the two functions more separate.
What This Means for OpenAI's Safety Approach
The move to combine research and safety under a single leader suggests that OpenAI wants to embed safety considerations more directly into the development process. By having Mia Glaese oversee both areas, the company may be aiming to reduce friction between teams that sometimes have competing priorities — one pushing for faster innovation, the other for more caution.
However, critics may argue that merging safety into the research chain could weaken independent oversight. The effectiveness of this new structure will depend on how much authority Glaese has to pause or redirect projects if safety concerns arise.
Our Take: A Risky but Necessary Reorganization
In our view, OpenAI's decision to merge safety and research teams is a double-edged sword. On one hand, it makes sense to have safety experts working closely with researchers from the start — rather than checking work after it is done. This could lead to safer AI systems being built more efficiently.
On the other hand, removing a separate safety leadership role could reduce the independence of safety assessments. When the same person is responsible for both building and checking, there is a natural conflict of interest. The real test will be whether Mia Glaese and her team have the authority to say "no" when needed — even if it slows down progress.
For now, the departure of Johannes Heidecke marks the end of an era where OpenAI's safety function had its own dedicated leader. The company is betting that integration will work better than separation. Only time will tell if that bet pays off.
