OpenAI's chief scientist Jakub Pachocki published an essay on 6 September titled An Alien Mind, in which he argues that frontier AI systems are becoming something humans do not fully understand, and that the current pace of progress could be sustained into recursive self-improvement, where the models increasingly drive their own development.
Pachocki describes machine intelligence as grown rather than designed, the product of repeating one straightforward optimisation step across enormous amounts of compute. The result, he writes, is a system that works through abstract concepts and behaves in ways that evade a full description, closer to an alien intellect than to human reasoning.
The essay is unusually direct about safety. Pachocki claims that GPT-6 Astra is significantly better aligned than GPT-5.6 Sol, but adds that no lab has yet solved alignment and monitoring well enough to keep scaling at maximum speed for much longer. He expects voluntary slowdowns to become commonplace until shared safety bars are agreed.
He also flags a weakening of OpenAI's main monitoring tool. The company bets on chain-of-thought monitoring, watching how a model reasons before it acts. Pachocki says that oversight is progressively diminishing, because models now blend reasoning with tool use and grow smarter even without verbalising their steps.
On security, Pachocki argues that today's models are becoming superhuman at breaking in and out of computer systems, and that there is a narrow window to harden critical infrastructure. He proposes safety commitments enforced by third party auditors, governments and international bodies, and calls international coordination on AI development a top priority for governments.
For founders, the practical signals are procurement and compliance. If mandated safety bars and audits spread, model providers will have to demonstrate alignment and monitoring to sell into regulated markets, and the price and availability of the most capable models may shift with them. Building a product on a single frontier model carries rising policy risk.