OpenAI is making a major reshuffle of its research and security teams Johannes Heidecke, Head of Security Systems is leaving the company. Heidecke shared his decision to leave directly with OpenAI employees, according to internal company memos seen by WIRED. After this, the company appointed its security teams to those who had previously managed this field. Saachi Jain and researcher Mia Glaese repositioning it around. Changes, OpenAI GPT-5.6 It is also important because it comes right after the launch of the model.
According to the information on his LinkedIn profile, Heidecke joined OpenAI in 2021 and took on various responsibilities in security efforts. He took over as President of Security Systems in 2024 following the departure of Lilian Weng. The technical documentation published by OpenAI for the GPT-4.5 and o1 models also shows Heidecke’s contribution to security evaluations and model testing. The company has not yet announced the exact date of Heidecke’s departure or his next role.
OpenAI will manage the security systems team during the transition process. To Saachi Jain will give. Jain comes to the role with internal and direct experience, having previously led security teams within OpenAI. However, the company keeps this appointment temporary for now and does not share details about the permanent manager. Thus, Jain will continue the daily work of security systems and ongoing model evaluations after Heidecke.
However, OpenAI supports all security teams. To Mia Glaese, Vice President of Research and Security will tie. Glaese has a long-standing involvement in the company’s research on secure model behavior, model alignment methods, and security policies. OpenAI’s studies on rule-based rewards and deliberative alignment are also among the topics to which Glaese contributed. With the new task distribution, Glaese will bring together research teams and product safety studies on the same management line.
OpenAI will involve security teams in model development earlier
OpenAI Research Director Mark ChenIn his statement to WIRED, he said that they want to combine security studies with the advanced model development process. Chen emphasized the importance of security teams being involved earlier and more directly in key model, product and release decisions. Accordingly, security experts will not only take part in final checks, but will also participate in model development studies from the first stages. With this task distribution, OpenAI will make security decisions closer to research and product processes.
Task changes, OpenAI GPT-5.6 model family It appeared right after the period when it was put into use. Initial reports suggested that the US government had given official approval to the model, but the White House later clarified that private companies did not need federal approval to release models. The government lifted restrictions on wider distribution after additional testing and discussions with OpenAI, Axios reported. Therefore, the term government approval in the news describes the completion of security reviews rather than an official licensing process.
On the other hand, OpenAI monitors the serious risks that advanced models may pose. Head of Preparation continues his duty. Sam Altman had previously explained through X that the company appointed this manager to pre-emptively identify and reduce serious risks. The preparation team is examining areas such as biological threats, cybersecurity and autonomous mobility of advanced models. The security systems team works on model behavior, product protections and pre-release security controls.