Shall we slow down?

Anthropic Institute (Marina Favaro, Jack Clark), early June — "We believe it would be good for the world to have the option to slow or temporarily pause frontier AI development to enable societal structures and alignment research to keep up with the advance of the technology." [source]

"Pacing the Frontier", a letter from frontier-lab employees, July 28 (1,386 signatures as of today, including Amodei, Kaplan, Clark, Pachocki) — "We request that the U.S. government support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development." [source]

OpenAI, August 18 — "a two-week pause in reinforcement learning (RL) training on our latest models intended for deployment… Our largest planned frontier RL run remains on hold." [source] Sam Altman: "We have paused some frontier RL training to ensure that we can meet the appropriate alignment, security and monitoring standards for the new level of capabilities in front of us." [source]

Sam Altman, David Senra's podcast, August 23 — "The economy just has so much inertia… I'm grateful for it. But I think it means we've all been too ambitious on timelines, even with this incredible technology… Society and the economy will adapt more slowly." [source]

Jakub Pachocki, OpenAI Chief Scientist, essay "An Alien Mind", September 6 — "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." [source]

Sam Altman, X, September 8 (the day OpenAI announced its Navier–Stokes result) — "We have been talking a lot about the need to pace progress to ensure safety; for me this is the strongest evidence yet of the urgency." [source]

Dario Amodei, essay "We Must Pace the Frontier", September 12 — "We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast, and we must make wise use of the time we gain." And: "pacing does not mean halting model training or technical progress, but ensuring companies take adequate time to align." [source]

Reactions within 48 hours — Altman: "I agree with Dario that we need to pace the frontier." [source] On OpenAI's IPO: "I would say not 2026." [source] Musk: "Dario is right." [source] Hassabis: "The details need working through, but the direction is correct for meeting this critical moment." [source] Nadella: "deliberate pacing needed to get alignment right." [source]

Bernie Sanders, letter to Altman, Amodei and Zuckerberg, August 10 — "Pause AI development. It is not too late to avoid disaster. Stop building machines that humans cannot control… If you do not take appropriate action now, my colleagues and I in the U.S. Senate will." [source]

Sanders + Greg Casar, the "Ban Artificial Superintelligence Act" bill, September 3 — a permanent ban on developing and deploying systems that match or exceed human cognitive performance; up to 20 years in prison (modelled on the penalty for unlawfully developing nuclear weapons) plus a "corporate death penalty"; a pause on advanced AI development until a cabinet-level regulator is established. [source]

Also, the very same Elon Musk, September 14 — "Grok 4.7 should be roughly on par with Opus 5.0, not 5.1. Better in some ways, worse in others. We need to fix multimodal performance. Grok 4.8 will be a noticeable improvement. Grok 4.9 is probably Astra/Fable class. Grok 5 maybe better than anything. We shall see." [source]

Why is it "impossible"? Simply put — it can't be enforced.

There's no way to verify that anyone has actually slowed down. Anthropic said so itself in June: "Training runs are far easier to conceal than missile silos, their inputs are general-purpose, and the incentive to defect quietly is enormous." It points to the INF Treaty, whose verification regime "took decades to build both the infrastructure and the trust" — and adds: "We don't have that long." [source]

Open models can't be recalled. Open models aren't going away, and you can't un-invent what you already know. The cat's out of the bag. It's too late.

China won't slow down, and without China a pause makes no sense. Even if US labs slow down, what stops everyone else from not slowing down? Amodei himself admits China is the "toughest dilemma" for his plan. [source]

The labs are caught in a competitive trap, and they admit it. From the "Pacing the Frontier" letter, signed by Amodei and Pachocki among others: "each company—and country—is under intense competitive pressure not to unilaterally slow that acceleration." [source]

The cost is collapsing. Noam Brown (OpenAI): "when @OpenAI announced o3 it cost ~$500,000 to score 87.5% on ARC-AGI 1. Today, Astra scores higher for ~$20." [source] A slowdown at a handful of labs doesn't bend that curve.

And the punchline.

TIME, September 15 — "no leading AI company has truly slowed down yet."