What happened
Johannes Heidecke, head of OpenAI's Safety Systems team, departed as the company reorganised its safety and research functions. Mia Glaese now serves as Vice President of Research and Safety, integrating safety deeper across research teams. An OpenAI spokesperson stated safety decisions require understanding model capabilities, and research decisions need safety implications. This follows multiple high-profile departures from OpenAI's safety-focused teams, including Jan Leike and Ilya Sutskever in May 2024.
Why it matters
Centralising safety under research leadership shifts the mechanism for embedding risk mitigation into model development. For platform engineers and security architects, this reorganisation creates a new internal structure for safety decisions, potentially impacting the transparency and independence of safety evaluations. This follows the dissolution of the Superalignment team in 2024 after its co-leads departed, indicating a persistent challenge in retaining dedicated safety expertise.
