Pachocki says no lab has solved AI alignment, urges slowdowns

OpenAI's chief scientist says no lab, including his own, has solved AI alignment well enough to keep scaling at maximum speed for much longer.

In an essay titled *An Alien Mind* published on September 6, Jakub Pachocki writes that 'no lab has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer.' He follows it with a direct policy ask: 'I expect and hope for voluntary slowdowns to become commonplace until shared safety bars are established,' and argues that 'international coordination on future AI development needs to become a top priority for governments around the world.'

Modern AI, Pachocki argues, is not built the way software is built. It is grown, and the result is an intellect humans do not fully understand.

The specific worry is monitoring. OpenAI has leaned on chain-of-thought monitoring, reading a model's step-by-step reasoning as a safety check, and Pachocki concedes the technique's effectiveness is progressively diminishing. Reasoning models now operate in far more complex environments, they are getting better at reasoning about their own reasoning, and stronger pretraining lets them reason without verbalizing a chain of thought at all.

He traces the concern to the summer of 2023, when he and a colleague, Szymon, watched the first results from an internal OpenAI project called RLSlow. His stated expectation now is that current progress could sustain into recursive self-improvement, with AI doing meaningful AI research on itself, within a few years.

'The idea of racing forward at all costs seems absurd once one internalizes the seriousness of the stakes,' he writes.

Eleven of the researchers we track shared the essay within the same cycle.