1. Home
2. News
3. AI safety

OpenAI's chief scientist joins Hinton and Bengio in telling governments to prepare for an intelligence explosion

A speed limit on capabilities growth is on its list of options, and OpenAI's chief scientist co-signed it.

By Think Facility · September 29, 2026 · AI safety AI policy

On September 28, 2026, a long list of researchers published a paper telling governments to get ready for an intelligence explosion. The Centre for the Governance of AI hosts it under the title What If Automating AI R&D Triggers an Intelligence Explosion?

Who signed it matters more than the title. Hinton and Bengio, sure. But OpenAI's chief scientist, Jakub Pachocki, is on it too. So are Microsoft's chief scientific officer Eric Horvitz and Anthropic co-founder Jack Clark. So some of the people running this research have put a speed limit on it among the options.

Published

September 28, 2026

Anthropic, August 2026

26% of internal AI R&D work with only high-level supervision

Five months earlier

1%

OpenAI, September 2026

R&D tasks that would take days for staff

Why they think it could happen soon

It comes down to headcount. A few thousand people do frontier AI research today. Copy an AI researcher and run it in parallel, the authors write, and you've added the equivalent of millions more, who then build their own better successors.

Their evidence is mostly the labs' own numbers. Anthropic's 26% is the figure we wrote up when it came out. Some experts, the authors say, think AI R&D could be fully automated within a few years (they don't name them here).

What they want governments to do

First, just find out what's happening. They want embedded auditors inside frontier companies, and they want those companies made to report how much of their R&D is automated and how fast the models are getting better.

The next part surprised me a bit, given who signed it. There's the option to shut down high-stakes experiments, and research kept on air-gapped networks. And there's this:

International agreements to pace progress are on the same list. Dario Amodei asked for much the same in his essay earlier this month. The last group is about getting society ready, with emergency plans for a major AI-driven cyber or bio incident and for losing control of the systems themselves.

What the authors said

Sam Manning, a co-author, announced it on X in a short thread. Alan Chan led it, he says (Chan's name is first on the list, too).

Among the risks the authors list is superintelligent systems escaping human control and acting to marginalize humanity. The summary gives no date for any of it.