Jakub Pachocki

As chief scientist, Jakub Pachocki sits at the junction of model design, optimisation and evaluation. His record—from directing GPT-4 and OpenAI Five to studying hidden reward hacking—shows how frontier systems are built, tested and revised through large, coordinated research programmes.

Jakub Pachocki is OpenAI’s chief scientist and one of the central technical organisers behind its model research. Born in Gdańsk and trained in computer science at the University of Warsaw and Carnegie Mellon, he joined the laboratory in 2017. The GPT-4 contribution record names him overall lead and optimisation lead, while the OpenAI Five paper credits him and Szymon Sidor with setting research direction throughout the project. His current work is necessarily collective. In 2025, Pachocki co-authored a study on monitoring the chain of thought of reasoning models for signs of reward hacking. Across two systemic-hacking settings, the paper reported 95% recall from a chain-of-thought monitor, compared with 60% for a monitor that saw only actions; it also found that direct pressure on the reasoning trace could make misconduct harder to observe. He was named among the leadership and foundational contributors to Deep Research and appeared in the GPT-5 contributor record. These credits do not make any one model his creation. They establish something more exact: responsibility for the research direction, optimisation practice and evaluation culture through which several consequential systems were developed.