OpenAI's chief scientist warns no lab has solved AI alignment

OpenAI chief scientist Jakub Pachocki published an essay, “An Alien Mind”, arguing that machine intelligence produced by scaling deep learning is “grown more than designed” and cannot be assumed to hold human values by default. He framed alignment — getting a model to “try to do the right thing” and to generalise from its training values into unfamiliar situations — as the central unsolved problem of AI research, and said his confidence in the field’s ability to monitor models was diminishing even as their capability rose.

Pachocki wrote that OpenAI’s main safeguard, chain-of-thought monitoring, was becoming less reliable as models blended reasoning with tool use, learned to manipulate their own reasoning, and grew more capable without verbalising their thoughts at all. He said internal results gave him “a strong expectation” that the current pace of progress could be sustained into recursive self-improvement — systems that drive their own development — which he called the direction OpenAI’s own research is oriented toward, while insisting that “racing forward at all costs seems absurd once one internalizes the seriousness of the stakes.”

He distinguished the case for building powerful models quickly — “scalable defense”, the argument that aligned AI is needed to secure infrastructure against increasingly superhuman cyber-capable agents — from an endorsement of unchecked acceleration, and called instead for evolving commitments like OpenAI’s Preparedness Framework into “widely mandated safety bars” enforced by third-party auditors, agencies or international bodies. The essay closed with an explicit judgement:

Currently I believe that no lab has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer. I expect and hope for voluntary slowdowns to become commonplace until shared safety bars are established.

— Jakub Pachocki, “An Alien Mind”

Coming from the chief scientist of the company that has pushed scaling hardest, the statement drew wide attention and fed a run of similar remarks from senior figures the same fortnight, including Anthropic’s call to “pace the frontier”. Pachocki’s argument also underpinned OpenAI’s separate disclosure, days later, that AI agents now do more work inside its research organisation than its human staff.