OpenAI’s organizational restructuring is quite interesting—the model training is getting faster, product releases are getting more urgent, and security coordination is becoming harder instead. Putting security into the research framework—hopefully it isn’t meant to let compliance make way for iteration speed.

Crypto Globe news: Johannes Heidecke, the former head of OpenAI’s safety system, has announced his departure to employees. At the same time, OpenAI is adjusting the structure of its safety team, further merging the safety team with the research team. Mia Glaese, former research vice president and head of alignment, will be promoted to vice president of research and safety, responsible for coordinating related work, and all safety teams will report to her going forward. Saachi Jain, who previously led the safety team, will temporarily serve as head of the safety system and report to Glaese. Mark Chen, OpenAI’s chief research officer, said that because the frequency of training models has accelerated significantly and the product release cycle has also shortened markedly, safety coordination is more difficult than before. They hope that with the new organizational structure, the safety team can be involved earlier in model, product, and release decisions.