OpenAI's Chief Scientist Published an Essay Saying No Lab Has Solved Alignment Well Enough to Keep Scaling at Full Speed
Jakub Pachocki published 'An Alien Mind' on September 6, 2026, writing that no lab has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer, and that he expects and hopes voluntary slowdowns become commonplace until shared safety bars exist. He splits the problem into goal alignment (does the system pursue the objective it was given) and value alignment (does it generalize principles where instructions run out), and says internal results give him a strong expectation that the current pace sustains into recursive self-improvement. Coming from the person running OpenAI's research org in the same week the company declared an automated research intern milestone, this is the sharpest public hedge any frontier lab scientist has put on their own roadmap.
Source
↳ Follow the thread