
A dated log of public calls to pause, slow or pace frontier AI, from Amodei and Hendrycks to the Pacing the Frontier statement, with what each asks and what the evidence supports.
This tracker logs the public calls to slow, pace or pause frontier AI development, dated, with who made them, what exactly they asked for and what the evidence says. Calls differ widely in what they would stop, and most coverage blurs that. We update this page when a new call or response is verified, and we link each entry to the document or the report we opened. Last updated November 22, 2026.
How to read the entries
Three separate asks get called "a pause." A pause on the largest training runs, such as Dan Hendrycks's proposal, is the strictest. A slowdown in the pace of capability gains, which is what Dario Amodei's essay asked for, is looser. A call for tools that would give governments the option to pace the frontier, which is how the Pacing the Frontier statement is worded, is looser still. The log below keeps them apart.
The log
September 12 to 14, 2026: Dario Amodei, Anthropic CEO. IAPP, publishing on September 14, reports Amodei's open essay urging companies to slow frontier development. It quotes him: "We must slow the pace at which we improve the capabilities of AI models." He said even "an extra year or two" could give time to improve alignment. His plan, as IAPP summarizes it, calls for embedded third-party evaluators, more collaboration among frontier developers and coordinated safety standards among democratic countries, aiming eventually at international agreements.
September 2026: lab and platform responses. IAPP reports Sam Altman backed the plan and said OpenAI will also embed independent evaluators; Demis Hassabis said "The direction is correct for meeting this critical moment," though some elements need reassessment; Elon Musk supported the proposal; Satya Nadella backed independent evaluators but said benefits must be broad. None of the reports we opened shows a verifiable cap on training runs from any of them.
September 13, 2026: Donald Trump. IAPP quotes the US president as opposing a slowdown: "whoever wins AI, wins." China's foreign ministry spokesperson Guo Jiakun, per the same report, said China has no plans to slow its AI development.
“We must slow the pace at which we improve the capabilities of AI models.”
Undated on the page when we read it, listed November 22: "Pacing the Frontier" statement. The statement's site lists 1,386 employees of frontier AI companies as signers. It says AI could help create a dramatically better future but that outcome is not guaranteed, that "the world lacks the technical and governance tools to deliberately pace frontier-wide progress," and that each company and country is under intense competitive pressure not to unilaterally slow. It asks the US government to support an international effort to develop the tools needed to pace the frontier of automated AI development. It is organized by employees with support from the nonprofits Guidelight AI Standards and Encode AI. It does not ask for a halt. Defense One, on November 19, cited a figure of more than 1,300 signers, which is consistent with a count that rose over time.
November 5, 2026: METR assessment, the evidence behind the urgency. METR's report on AI R&D acceleration at Anthropic estimated Anthropic is improving models on well-specified tasks at about twice its earlier rate, and that about 80% of the current acceleration is due to AI, a figure METR calls highly uncertain. Anthropic redacted a majority of the evidence METR found most informative. Defense One reports that Sentinel now expects superhuman AI around 2028, not 2031, and that the AI Futures Project's November 10 update says superhuman AI will "likely arrive at some point in 2027" absent deliberate slowdown. These are forecasts, not measurements.
November 2026 (reported November 21): Dan Hendrycks, Center for AI Safety. At the New York Times DealBook Summit, per The Next Web, Hendrycks urged an indefinite pause on the largest training runs, with carve-outs for narrow models and inference sales allowed to continue. Max Hodak said enforcement would require a surveillance state. We cover the panel in a separate news report.

The counter-evidence and the open problems
Enforcement is the central dispute. Pause advocates point to the physical size of frontier training clusters; critics, including Hodak, argue monitoring would reach far into the internet. China's stated position, per the IAPP report, is that it will not slow its AI development, which any pause proposal must answer. Verification of training runs has no agreed method in the documents we opened.
There is also litigation: our tracker of lawsuits to slow frontier AI covers the court route, which is separate from the calls logged here.
What each side would need to show
For a pause to become policy, proponents would need a verification method, a position on non-US labs and a legal vehicle. For opponents, the burden is to say what would count as a warning sign serious enough to slow down, given that several forecasters and METR's own assessment are moving timelines earlier. Neither side has published that in the sources above.
How we will update this page
Entries are added when a call, a response or a piece of evidence is verified against a document we opened, and each carries its date. We do not log rumors, anonymous claims or social posts we cannot source. When a signer count or a quote changes we will note the change with a date instead of editing silently, in line with our corrections policy.
Our take
The strongest evidence in this log is not the rhetoric. It is the METR finding and the redactions around it: an independent assessor saying acceleration is real and that most of the evidence cannot be checked by the public. We will keep adding dated entries, including any bill text, any lab commitment on verification and any change in the positions above, and we will date each correction when a figure or quote is revised.
Frequently asked questions
Who has called for a pause on frontier AI?
Dan Hendrycks called for an indefinite pause on the largest training runs at DealBook, reported November 21, 2026. Dario Amodei called in September 2026 for slowing the pace of capability gains, which is a looser ask.
What is Pacing the Frontier?
A statement listing 1,386 employees of frontier AI companies. It asks the US government to support an international effort to develop tools to deliberately pace frontier AI, and does not ask for a halt.
Did the AI labs agree to slow down?
IAPP reports Altman, Hassabis, Musk and Nadella backed parts of Amodei's plan, mainly independent evaluators. The reports we opened show no verifiable cap on training runs from any of them.
Would China agree to a pause?
China's foreign ministry spokesperson Guo Jiakun said China has no plans to slow its AI development, according to IAPP's September 14, 2026 report.
Why do forecasters say timelines are shorter?
Defense One reports METR's November 5 assessment of AI R&D acceleration at Anthropic led some to shorten timelines, with Sentinel now expecting superhuman AI around 2028. These are forecasts, and Anthropic redacted much of the supporting evidence.
Sources
What each one is, and whose it is.
- 1
AI leaders support call to slow pace of frontier development as safety concerns grow, IAPP (September 14, 2026)
Press reportIndependent of the vendor - 2
Pacing the Frontier, Employees of frontier AI companies (November 22, 2026)
OtherIndependent of the vendor - 3
Independent assessment of AI R&D acceleration at Anthropic in October 2026, METR (November 5, 2026)
OtherIndependent of the vendor - 4
Dan Hendrycks calls for a pause on frontier AI, Max Hodak pushes back, The Next Web (November 21, 2026)
Press reportIndependent of the vendor - 5
'I was seriously wrong': Experts say superintelligent AI may arrive as soon as next year, Defense One (November 19, 2026)
Press reportIndependent of the vendor