Pachocki warns that AI scaling is outpacing safety readiness

In “An Alien Mind,” OpenAI chief scientist Jakub Pachocki argues that no lab has sufficiently solved alignment and monitoring to sustain maximum-speed scaling much longer. He calls for voluntary slowdowns, common safety requirements and international coordination. Jakub Pachocki / OpenAI 

Citing internal results, he expects AI to increasingly contribute to its own development. He reports declining confidence in monitoring models through their written reasoning, and warns that trained values may not hold in unfamiliar situations. These are his assessments, not independently established findings in the essay. Jakub Pachocki / OpenAI 

Pachocki supports using AI for alignment research and defense, while constraining further scaling by safety confidence. He proposes enforceable safety requirements overseen by independent auditors, governments or international bodies, and says OpenAI will withhold further scaling when needed. Jakub Pachocki / OpenAI