Johannes Heidecke
Johannes Heidecke worked on AI safety at OpenAI and led its Safety Systems team. In 2018, he worked on inverse reinforcement learning at AI Safety Camp and coordinated its Prague edition. He joined OpenAI as an AI safety analyst in 2021, co-authored the GPT-4 Technical Report in 2023, and took over Safety Systems in 2024. His later research included Deliberative Alignment, which trained models to reason over safety specifications; in April 2025, he led Safety Systems and focused on AI safety and reinforcement learning. In July 2026, he told staff he was leaving OpenAI as the company reorganized its safety and research teams.
Timeline
-
2018
AI Safety Camp work
He works on an AISC1 team studying human irrationality through inverse reinforcement learning in April 2018, then coordinates AISC2 in Prague in October 2018.
-
2021
Joining OpenAI
He joins OpenAI as an AI safety analyst in 2021.
-
March 2023
The GPT-4 report
He co-authors the GPT-4 Technical Report submitted in March 2023, which describes the model’s development and post-training alignment.
-
2024
Leading Safety Systems
He takes over as OpenAI’s head of Safety Systems in 2024.
-
December 2024
Deliberative Alignment
He co-authors Deliberative Alignment in December 2024, a method that teaches models to reason over safety specifications before answering.
-
April 2025
Head of Safety Systems
He leads OpenAI’s Safety Systems team and focuses on AI safety and reinforcement learning in April 2025.
-
July 2026
Leaving OpenAI
He tells staff that he is leaving OpenAI in July 2026 as the company reorganizes its safety and research teams.
Sources and credits
Photo: We have no published photo of Johannes Heidecke yet. Send us a free-licensed photo.