Jakub Pachocki, Chief Scientist at OpenAI, warned that artificial intelligence is evolving at an extremely rapid pace, making it increasingly difficult for humans to understand and control. He expressed hope that AI labs would voluntarily slow down development for safety considerations.

"It Won't Be Long Before Models Can Improve Themselves"

In an article published on Sunday, Pachocki wrote that in a relatively short time, AI models have already been able to operate computers, collaborate with humans and other AI systems, and conduct research projects. He stated that soon, these models will also be able to improve themselves without human intervention, a process known as "recursive self-improvement."

"This is a moment that requires extreme caution," Pachocki wrote. "I worry that no one is prepared for the consequences of machine intelligence continuing to improve rapidly."

Voluntary Slowdown Should Become the Norm Before Establishing Common Safety Thresholds

He wrote that developers could try to guide AI more strictly to align with human interests, or "slow down future development when necessary." He said, "I expect and hope that voluntary slowing down of development will become common practice before establishing common safety thresholds."

This statement aligns with recent data disclosed by OpenAI: while the company is using agents to accelerate AI development, it also acknowledges that it cannot assume alignment and safety capabilities will keep up with model capabilities. More powerful systems may be harder to monitor. Pachocki previously stated that no laboratory has yet made AI alignment and monitoring sufficiently reliable, and he called on governments to prioritize international coordination.