← Back to the dossier

September 2026 · OpenAI, Anthropic, Google DeepMind, xAI

The people building it start asking for a slowdown

OpenAI's chief scientist wrote that no lab has solved alignment well enough to keep scaling at full speed. Sam Altman, Dario Amodei, Demis Hassabis and Elon Musk all said they would support a slowdown in frontier development. That is the four largest labs, in public, in the same month.

ArgumentCorroboratedMore than one independent source, or a primary document.

What happened

Jakub Pachocki, OpenAI's chief scientist, published an essay on 6 September on increasingly capable AI and the research still needed to keep systems aligned.

The statements from the four executives came after Hugging Face and the run of disclosures that followed it. They are statements of support for a slowdown, not a slowdown. OpenAI did actually slow its own training; what the others did is a separate question this entry does not answer.

What it points at

Read the Pachocki quote twice. It is the chief scientist of the company that built the model in the July incident saying, in writing, that nobody has solved this, that he hopes voluntary slowdowns become common until shared safety standards exist, and that international coordination should be a top priority for governments.

That last clause is a request aimed at people like the ones in a room in Leuven. It is also the thing Europe is currently not doing: see 7 May.

The cynical reading is available and worth stating: a lab calling for regulation after an incident it caused is also a lab shaping the regulation it will get. Both readings can be true.

Editor's notewhat we make of it, kept apart from what happened

The Pachocki quote is the single most useful thing you can read out loud all evening. It is long. Read it anyway, slowly.

Then give the cynical reading, because someone in the room is already thinking it and it is better coming from you.

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. And I believe that international coordination on future AI development needs to become a top priority for governments around the world.
Jakub Pachocki, chief scientist, OpenAI · OpenAI timeline, entry of 6 September, on the Pachocki essay

Sources

  1. OpenAI timeline, entry of 6 September, on the Pachocki essayprimary · main sourceThe quote is given in full there. Link the essay itself once you have its URL.
  2. Axios: OpenAI agents tried hacking various sites in May, JunepressFor Altman, Amodei, Hassabis and Musk saying they would support a slowdown.

Read next