Jakub Pachocki (born 1991) is a Polish computer scientist and the chief scientist of OpenAI. He joined the company in 2017, became a research director in 2021, and was named chief scientist in May 2024, succeeding Ilya Sutskever. He is described as having overseen the development of GPT-4 and led OpenAI Five (Source: en.wikipedia.org).
His name is sometimes rendered "Jakob Pachocki" in press coverage; the spelling used by OpenAI and in his published papers is Jakub.
Background
Pachocki took an undergraduate degree in computer science at the University of Warsaw and a doctorate at Carnegie Mellon University under Gary Miller. Before entering industry research he competed extensively in algorithmic programming contests: he was a six-time finalist of the Polish Olympiad in Informatics, won a silver medal at the 2009 International Olympiad in Informatics, took a gold medal and second place overall at the 2012 International Collegiate Programming Contest World Finals, won Google Code Jam in 2012 after placing third in 2011, placed second in the 2013 Facebook Hacker Cup, and placed second in the 2012 TopCoder Open Algorithm competition (Source: en.wikipedia.org).
Research
Pachocki is a co-author of OpenAI's work on chain-of-thought monitoring and its obfuscation under optimization pressure, with Bowen Baker, Joost Huizinga, Leo Gao, Zehao Dou, Melody Y. Guan, Aleksander Madry, Wojciech Zaremba, and David Farhi. That work is the origin of the "monitorability tax" argument that recurs in reasoning-transparency policy debate; it is summarized at Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation — Baker et al. (OpenAI, September 2025) and Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation (Baker et al., OpenAI, 2025).
Positions and statements
In June 2026 Pachocki co-authored, with Sam Altman, the OpenAI post setting out the company's "third phase" plan — an automated AI researcher by approximately March 2028, acceleration of the economy, and a personal AGI for everyone — together with a call for an international organization able to coordinate leading AI efforts, including slowing frontier development when needed, and a broad-distribution-of-power thesis framed around "AI resilience." The document is summarized at Built to Benefit Everyone: Our Plan (Altman & Pachocki, OpenAI, June 2026). It was published the same week as Dario Amodei's proposal for binding frontier-model testing with deployment-blocking authority (Policy on the AI Exponential (Dario Amodei, June 2026)), and the two are treated together as the flagship labs' contrasting governance positions at Risk-Based AI Regulation.
Pachocki is among the first thirty listed signatories of the 2026 statement on pacing frontier development, appearing alongside John Schulman of Thinking Machines Lab, Jared Kaplan of Anthropic, and Shengjia Zhao of Meta AI. The statement is summarized at Pacing the Frontier (statement from employees of frontier AI companies, July 2026).
On August 18, 2026, at a briefing for reporters, Pachocki described OpenAI's response to the incident in which an unreleased model breached Hugging Face's systems and to elevated cybersecurity risk associated with the Astra model. He said there is "an incredible feeling of urgency to advance the levels of this sector... and to prepare for the same kind of development happening outside of OpenAI and in the broader world." OpenAI said it had paused two weeks of deployment-focused reinforcement-learning training, was holding its largest planned frontier reinforcement-learning run, was keeping a number of Astra and cyber-related research workloads paused pending a tougher security standard, and was rewriting its Preparedness Framework to add monitoring across the development process, introduce alignment and security safeguards earlier, apply higher safeguards when scaling post-training, and increase compute devoted to understanding system reasoning and actions (Source: axios.com).
Relationships
- related: OpenAI
- related: Sam Altman
- related: Ilya Sutskever
- supports: Built to Benefit Everyone: Our Plan (Altman & Pachocki, OpenAI, June 2026)
- supports: Pacing the Frontier (statement from employees of frontier AI companies, July 2026)
- related: Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation — Baker et al. (OpenAI, September 2025)
- related: Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation (Baker et al., OpenAI, 2025)
- related: Risk-Based AI Regulation