OpenAI’s Chief Scientist Called for Slowing Down AI Development

• OpenAI chief scientist Jakub Pachocki acknowledged that it may be necessary to slow the development of advanced AI systems.
• In his view, existing alignment and monitoring mechanisms are still insufficient to keep scaling models at maximum speed.
• He proposed making safety standards mandatory across the entire industry.
• Similar initiatives were previously put forward by Anthropic CEO Dario Amodei and Google DeepMind head Demis Hassabis.

OpenAI chief scientist Jakub Pachocki urged AI companies to be prepared to slow the development of the most powerful models if existing safety mechanisms do not allow them to confidently control their behavior. He wrote about this in the essay An Alien Mind.

According to the researcher, artificial intelligence is beginning to play an increasingly significant role in its own development. If current rates of progress continue, this process could shift into recursive self-improvement, where AI directly accelerates the creation of more advanced systems.

Pachocki believes the industry needs to improve model alignment and monitoring mechanisms in parallel, while also being ready to limit further scaling when safety assurances are insufficient.

He noted that no lab has yet reached a level of control that would allow it to keep ramping up AI capabilities at maximum speed over the long term. At OpenAI, he said, they are prepared, if necessary, to independently refrain from further scaling of certain systems.

Pachocki singled out the declining reliability of monitoring chains of reasoning as a separate problem. More advanced models are better at analyzing their own thought process, while some of their capabilities may manifest without explicit, verbalized reasoning. This makes oversight by developers more difficult.

As one example, he cited a recent incident involving OpenAI agents. As we previously reported, during internal tests the systems created unauthorized communication channels, coordinated actions, and gained access to the infrastructure of Hugging Face and OpenAI itself.

After that, the company began developing a mechanism to automatically shut down AI systems when dangerous behavior is detected. 

AI Companies are Increasingly Talking About Slowing Development and Tightening Oversight

Paczocki believes that voluntary commitments by individual labs will not be enough in the long run. In his view, mechanisms such as OpenAI’s Preparedness Framework and Anthropic’s Responsible Scaling Policy should be turned into shared, mandatory standards. Compliance could be overseen by independent auditors, government agencies, or international organizations.

Similar proposals are being voiced more and more often across the AI industry. In June 2026, Anthropic called for the creation of a global pause mechanism that would make it possible to temporarily slow the development of the most powerful models when critical risks emerge.

Soon after that, the company’s CEO, Dario Amodei, argued for tighter government oversight. In particular, he proposed introducing mandatory safety requirements for the most powerful systems and strengthening independent industry monitoring.

In July, Google DeepMind head Demis Hassabis also called for the creation of a dedicated oversight body for frontier AI. The structure is expected to focus on standardization and independent model testing. Looking ahead, Hassabis also allowed for temporarily slowing development if risk levels become too high.

Overall, Paczocki’s proposal continues an emerging industry trend of moving from voluntary safety measures to more formalized oversight of the development of the most powerful AI systems.

Сообщение OpenAI’s Chief Scientist Called for Slowing Down AI Development появились сначала на INCRYPTED.