On September 6, 2026, OpenAI Chief Scientist Jakub Pachocki published an essay titled "An Alien Mind" on the company website. The core claim: no AI lab, including OpenAI, has solved alignment and monitoring well enough to justify scaling models at maximum speed.
What he said: Pachocki wrote that he expects and hopes voluntary slowdowns will become common until the industry establishes shared safety standards. He called for mandated safety thresholds enforced by third-party auditors, government agencies, or international bodies. He warned that OpenAI's main safety tool — watching a model's written reasoning, called chain-of-thought monitoring — is losing effectiveness as models get better.
Pachocki drew a distinction between goal alignment (does the AI try to do what you asked) and value alignment (does it still respect human values when the goal gets hard or unclear). He said current training methods can break under enough pressure, and that even aligned-seeming thoughts can bend toward the objective when optimization is pushed far enough.
The recursive loop: Pachocki revealed that internal results give OpenAI strong confidence that current progress could extend into recursive self-improvement — AI systems improving their own capacity to improve. He expects the next few years to deliver jumps as large as the last three, with AI increasingly driving its own development. Same day, OpenAI published data showing its researchers now use 3.1 agent-workdays for every human workday, spending over 600 dollars per day in inference at API prices.
Why it matters: This is not a rival or a critic. This is the person running research at the company that set the pace for the entire industry, publicly stating in writing that nobody has cracked the problem. He handed regulators a citable admission.