Coordinated slowdown proposals compared

Proposals to deliberately slow frontier AI development share a premise and diverge on almost everything else. The shared premise is a collective-action problem: no company or country will slow down alone, because the actor that continues inherits the lead. What separates the proposals is what they actually ask for — a moratorium, a standing option, a verification capability, a deterrence regime, or capacity-building — and how far each specifies triggers, adjudication, and verification.

The comparison below sets the proposals side by side; it does not take a position on whether any of them is desirable or workable.

The proposals

What each proposal actually asks for

The proposals fall into four groups by the strength of what they request.

Moratorium. The 2023 FLI letter asks for a bounded halt: at least six months, on systems above a stated capability reference point, with government moratorium as the fallback if labs do not pause jointly (FLI — Pause Giant AI Experiments: An Open Letter). The Ban Artificial Superintelligence Act announced by Sen. Bernie Sanders and Rep. Greg Casar in September 2026 was the first later proposal in this set to take that form, and goes further by pairing a pause pending agency rules with a permanent prohibition on a capability class; its text had not been released as of mid-September 2026 (Ban Artificial Superintelligence Act (announced 2026)). As described in coverage on September 16, the Sanders proposal threatens up to 20 years in prison for violations, a penalty its authors compare to those for illegally developing nuclear weapons (Source: axios.com).

Sen. Elizabeth Warren (D-Massachusetts) took the pause without the prohibition. Her office told Axios on September 16, 2026 that she would say that day, ahead of a roundtable hosted by Sanders: "We should immediately press pause on the development of advanced AI while lawmakers and regulators put systems in place to keep people safe," and "Congress must urgently pass legislation to put stronger guardrails in place before we have a cyberattack, economic crisis, or national security disaster facilitated by AI." She called the industry's push to "pace the frontier" "insufficient," arguing that elected officials rather than AI chief executives should decide the technology's future. The same account reports that former Vice President Kamala Harris had that week endorsed slowing down and called on President Trump to pursue a treaty with China, and that former President Barack Obama had backed slowing down as a first step (Source: axios.com). Neither Warren's call nor the Sanders–Casar announcement states a capability threshold for what the pause covers, and both tie its end to government action rather than to a test.

Rate limitation. Amodei's September 2026 essay is the first proposal from a frontier-developer chief executive to ask for slowing itself rather than for the option to slow. It distinguishes pacing from stopping, stating that pacing "does not mean halting model training or technical progress," and specifies a verification step, embedded third-party evaluators with employee-like access, which Anthropic committed to adopt unilaterally.

Standing option. "When AI Builds Itself" and "Pacing the Frontier" both ask for the capability to slow rather than for slowing. Favaro and Clark state the claim conditionally — "we believe it would be good for the world to have the option to slow or temporarily pause frontier AI development" — and argue that a unilateral pause "accomplishes much less" because it changes who leads rather than the underlying dynamics (When AI Builds Itself (The Anthropic Institute, June 4, 2026)). The 2026 employee statement makes the same move in one sentence, asking that the US government "support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development" (Pacing the Frontier (statement from employees of frontier AI companies, July 2026)). Neither names an instrument, a threshold, or a trigger.

Substitute mechanism. Superintelligence Strategy accepts the collective-action premise and answers it with deterrence rather than agreement: if each state can degrade a rival's destabilizing AI project, no state need trust the other to stop. Hassabis's framework routes the question through an industry standards body, where a slowdown is one escalation rung on a regime whose default output is pre-release testing.

The verification problem

The proposals that engage the mechanics converge on verification as the binding constraint, and on why it is harder for AI than for prior arms control. Favaro and Clark argue that training runs are far easier to conceal than missile silos, that the inputs are general-purpose, and that the incentive to defect quietly is large because whoever continues inherits the lead; a credible pause, they write, "has to specify what triggers it, what lifts it, and who adjudicates" (When AI Builds Itself (The Anthropic Institute, June 4, 2026)). They cite the Intermediate-Range Nuclear Forces Treaty as precedent while noting such regimes "took decades to build both the infrastructure and the trust."

This is why compute governance recurs across the set: compute is the input most amenable to counting, siting, and export control, and therefore the most plausible verification surface. It is also why the 2026 proposals ask for tool-building rather than for the tool's use — on their own account, the instruments do not yet exist.

A developer offered candidate measures on September 17, 2026. The Anthropic Institute proposed three quantities any frontier developer could publish — a R&D Automation Index of how much AI R&D work AI performs, coverage and escalation metrics for internal agents, and the share of AI R&D compute allocated to safety — and reported Anthropic's own values, stating that it "would expect these numbers to shift if there were coordination on pacing the frontier." It calls compute "among the most verifiable inputs to the AI R&D process" and therefore "a critical lever in a future pacing effort," and names possible uses of such measures: a trigger for "a fixed testing window before a new model is used for further AI R&D," commitments on the share of compute going to safety research, or "limits on the share of compute going towards AI research agents." The figures were produced largely by Anthropic's own models, and the post proposes third-party verification without reporting any (Measurements for understanding the pace of AI development inside frontier labs (Anthropic Institute, September 2026)).

Who signs, and what that indicates

The proposals differ in the standing of their supporters more than in their content.

The 2023 statements were signed largely by researchers and public figures outside the labs, or by lab leaders speaking about risk in the abstract. The 2026 employee statement inverts this: its signatories are employees of the companies whose acceleration is the subject, including the chief scientist or equivalent research lead of OpenAI, Anthropic, Meta AI, and Thinking Machines, four Anthropic co-founders, and Google DeepMind's chief strategy officer (Pacing the Frontier (statement from employees of frontier AI companies, July 2026)). The site listed 1,178 signatories at release on July 28, 2026 and 1,224 when observed the following day, and states that comments are made in a personal capacity and do not necessarily represent any company's views.

The comments published alongside the signatures span a wider range than the statement text. OpenAI's Leo Gao described "a deadly race towards an intelligence explosion" and wrote that "to survive, we must coordinate to slow down the race"; OpenAI's Joshua Achiam signed while stating that he does not know "what form such tools should take" and hopes any governance tools "will not be expansive or excessive." A signature on this statement is therefore weak evidence about a signatory's position on any particular instrument.

Countervailing positions

Two lines of argument run against the set.

The competitive-necessity argument holds that slowing is either impossible or self-defeating. Mark Zuckerberg made the sharpest recent version on July 28, 2026, writing that "invention, not automation, will be the greatest contribution of superintelligence," that America's "advantage is decentralized and open innovation," and that closing models to keep China from gaining access "will only disadvantage the US and its allies" (Source: wsj.com). Effective Accelerationism (e/acc) holds the position in its stronger form.

The safety-through-speed argument accepts the risk premise but concludes that safety-conscious labs should lead rather than slow — the position Anthropic has advanced alongside export controls and transparency legislation, and which sits awkwardly beside the same company's sponsorship of pause-verification research. Anthropic's own economic policy framework states the collective-action constraint plainly: "any firm that slows down alone simply cedes the market to those that do not."

The antitrust question

Every design surveyed here that relies on developers acting together runs into the same unresolved legal question, which a participant put to Congress directly in September 2026. OpenAI asked members of Congress in the weeks to September 10, 2026 for clear guidance on whether orchestrating an industry-wide slowdown on frontier AI development would be legal under antitrust law, according to people close to the company. Only the lede of that account was retrievable behind its paywall, so the form of guidance sought, the members approached and any response are not established (Source: wired.com).

The request is contemporaneous with the company's own proposal, published September 9, to build industry-led frontier monitoring standards with other laboratories "with or without government support" (The AI policy window is open. We need to act. (Chris Lehane, OpenAI, September 2026)), and with chief scientist Jakub Pachocki's statement that he expects and hopes voluntary slowdowns will become commonplace until shared safety bars are established (An Alien Mind (Jakub Pachocki, OpenAI, September 2026)). No safe harbor, business-review letter or statutory exemption addressed to coordinated AI safety measures is recorded here.

The same point had been made from inside the developers' research staff. Joe Benton, who left Anthropic's safety team in late August 2026 for METR, wrote on September 11, 2026 that he thought "even holding AI progress at today's pace, rather than the much faster pace the companies are aiming for, could be a win," that the technical machinery for it can be built, and that what is missing is "political will, and legal cover, since companies that agreed to restrain themselves together would currently risk antitrust liability." He paired this with a call for capability disclosure, incident and near-miss reporting, minimum-adequacy safety frameworks and independent assessment, arguing that political will for restraint is unlikely "while the frontier is invisible to everyone outside the companies" (Why I left Anthropic's safety team to hold AI companies accountable (Joe Benton, September 2026)).

A legislative vehicle already existed. Senators Adam Schiff (D-California) and Jim Banks (R-Indiana) had introduced the Collaboration on Adversarial Threats and Security Risks Act (S. 5105) on July 23, 2026, with an identical House bill, H.R. 9914; newsletter coverage in the week of September 14 carried it without a bill number or introduction date. Section 3(a)(2) exempts from the antitrust laws agreements among developers "for the exclusive purpose of reducing covered artificial intelligence security risks via delaying or otherwise limiting the release, deployment, use, development, training, testing, or evaluation" of AI, provided the parties first give written notice to the Assistant Attorney General in charge of the Antitrust Division. The coverage described the notice as running to the attorney general and the price-fixing safeguards as applying generally; in the text the Attorney General holds only the section 4 injunction power, and the rule of construction barring price-fixing and market allocation is drafted against the information-sharing limb, subsection (a)(1), alone (S. 5105 — Collaboration on Adversarial Threats and Security Risks Act (bill text); Source: washingtonpost.com). Seven House members joined H.R. 9914 as cosponsors on September 14 and 15, 2026, bringing it to 20 (Source: congress.gov). Paul Lekas, an executive vice president of the Software and Information Industry Association, said the coordination described in Amodei's essay "would go even beyond" existing legal comfort and that "you would need an explicit safe harbor." Against the safe-harbor approach, Doug Calidas of Americans for Responsible Innovation said that if laboratories "agree to restrict output, that's a pretty classic antitrust violation," and former FTC chair Lina Khan wrote on X on September 13, 2026 that laws already govern the release of dangerous, defective products, warning of the "highly concentrated and interconnected structure of these markets" (Source: washingtonpost.com). White House AI adviser David Sacks made the same product-liability point from the opposite direction, and in a September 12, 2026 post on X said observers should "stop pretending the motivation to slow down is purely altruistic" (Source: insideaipolicy.com). See AI and Tort Liability.

Altman narrowed OpenAI's own ask on September 14, 2026, saying in a social media post that the company does not believe it needs an antitrust exemption or legislation to begin the work, while welcoming "a federal framework that sets consistent safety requirements for frontier AI" (Source: washingtonpost.com; platformer.news). OpenAI global policy chief Chris Lehane said on September 15 that the company's safety engagement with Anthropic and Google DeepMind had been under way for several weeks and that he did not see the need for an antitrust waiver for the three to talk about safety; only the lede of that account was retrievable (Source: bloomberg.com).

The first public signal of the administration's view came the same day. Federal Trade Commission chairman Andrew Ferguson, stating that he spoke personally and not naming Anthropic, said at Georgetown University that "if companies are simultaneously coming to Washington and asking for a host of regulations and an antitrust exemption, all of my alarm bells go off," and that "they're asking for barriers to entry that will insulate their incumbency from challenge... everyone should be deeply suspicious about this." He said President Trump would set federal AI policy. Antitrust experts cited in the same account argued that existing doctrine already permits coordination to avert catastrophic risk, several pointing to a past statement by US antitrust authorities that cybersecurity information sharing is lawful. Cohere co-founder Aidan Gomez wrote in a September 13 blog post that the proposal was the "latest in a string of recent efforts by Silicon Valley incumbents to shape the regulatory landscape surrounding AI" and that "the rules for it cannot be written by a small group of commercially aligned companies behind an antitrust waiver" (Source: reuters.com). See AI Antitrust.

Congressional Democrats divided on the waiver. The letter from 107 House Democrats to Speaker Mike Johnson, dated September 15 and released September 16, 2026, lists "a waiver of antitrust laws to enable critical industry-wide collaboration on safety and security" among proposals that "deserve consideration, public debate, thoughtful amendment, and action," alongside transparency and evaluation mandates and kill-switch requirements (Letter from 107 House Democrats to Speaker Mike Johnson on AI safeguards (September 2026)). Sen. Mark Warner said in an interview on September 16: "Nobody would climb on an airplane or drive a car without some level of outside safety reviews. And the industry would need something like potentially even an antitrust exemption to talk to each other" (Source: reuters.com). Sen. Elizabeth Warren the same day called suggestions to weaken antitrust protections to confront AI risks "transparently self-serving"; the account of her remarks cites former antitrust officials who say an exemption would not be necessary because the law does not prevent AI companies from coordinating to prevent harm and sharing legitimate cybersecurity information (Source: axios.com).

Treasury Secretary Scott Bessent addressed a different exemption. Asked by Rep. Juan Vargas (D-California) at a House Financial Services Committee hearing on September 15, 2026 what the administration was doing on AI safety, he said the government "shouldn't do on safety is to give these labs a liability exemption, which is what they are asking for," that "the best way to guarantee safety is that the creators are liable for what they build and generate," and that labs were saying "'we would like to all slow down, but please give us a waiver on liability,' which should not be done" (Source: fedscoop.com). The waiver Amodei's essay requests is an antitrust waiver "for certain kinds of safety conversations" (We Must Pace the Frontier (Amodei, September 2026)); the same account notes that OpenAI had separately backed an Illinois bill limiting liability for AI-caused harms. See AI Liability. The next day Bessent said in a statement to Axios that the United States was "open to discussions on avoiding shared risks and avoiding bifurcation of our two systems" with China and expected talks with Vice Premier He Lifeng to "cover both open and closed weight models"; Axios set the statement against the administration's rejection of industry calls to pace or slow AI development (Source: axios.com). See US-China AI Competition: Different Races, Different Metrics.

The question reached a court on September 18, 2026. Four subscribers to ChatGPT, Claude, Grok and Gemini filed a proposed class action in the Northern District of California against Anthropic, OpenAI, SpaceXAI and Google, Buist v. Anthropic, PBC (No. 3:26-cv-10693), alleging that the four agreed to slow the pace at which they improve competing products in violation of Section 1 of the Sherman Act. The complaint pleads Amodei's essay as the offer and the September 12 statements of Musk, Altman and Hassabis as acceptances, cites OpenAI's inquiry to Congress, Altman's statement that OpenAI would not wait for an exemption and Lehane's account of weeks of prior collaboration as evidence, and reads the embedded-evaluator step as the means of confirming that each participant was slowing as agreed. It disclaims any challenge to unilateral safety measures or to the companies' advocacy for an exemption, and seeks treble damages for subscribers from September 12 and an injunction against horizontal agreements on the rate of development, compute or training-run limits, and capability checkpoints (Class action complaint, Buist v. Anthropic, PBC (N.D. Cal., No. 3:26-cv-10693, filed September 18, 2026); Source: broadbandbreakfast.com).

An industry safety body

Anthropic, OpenAI and Google were "privately discussing the creation of a new safety body" to coordinate safety standards, according to a Washington Post report of September 14, 2026 that credits The Information with first reporting the discussions. Only the lede of the Post account was retrievable, so the body's proposed form, membership terms and standing are not established (Source: washingtonpost.com). The discussions predate Amodei's essay. A fuller account, which became public on September 24, 2026, reported that the three companies were pushing forward with the body on their own, without government oversight, under the tentative name Standards Authority for Frontier AI (SAFA) and with a target launch at the end of 2026 or early 2027; according to people familiar with the matter, it would support third-party pre-deployment testing, set out how developers report safety and security incidents, define the laboratories' voluntary commitments in practice and set qualifications for independent auditors, while whether it would test models itself was still being debated. The account names candidates approached or considered for its leadership and scientific advisers (Source: theinformation.com); details are at AI Pre-Release Vetting.

A different industry-level mechanism was proposed at the All-In Summit in Los Angeles on September 14, 2026. Elon Musk said xAI, OpenAI, Anthropic, Google, Meta and "three or four of the leading Chinese companies" should let rivals run a "test harness" on their models before release: "instead of grading your own homework, you would at least have competitors grading your homework and raising the alarm if they see concerns." He acknowledged that no rival had agreed, said the method was not perfect but "the odds that you will find issues are dramatically greater," and said it was "probably something that China would agree to" (Source: cnbc.com). Peer testing by competitors addresses verification rather than pace; Amodei's essay places the same function with embedded third-party evaluators.

Positions of developers and their chief executives

Salesforce's Dreamforce conference in San Francisco on September 15, 2026 put the two sides on one stage. Amodei, speaking with Marc Benioff before about 12,000 people, said slowing development is "the way to lead the industry forward, to set an example, to say that everyone can always be better." Jensen Huang, who followed him, said market forces already exist and the industry needs no new laws or regulations, called the trade-off between speed and safety "a false choice," and put responsibility on developers: "Run as fast as you can. But if you feel at any given point in time the company's out of control or the product's not going to be safe, take a pause and make sure you get it right." Benioff told reporters that companies which feel they should slow down should do so, "and then they should be held accountable" (Source: cnbc.com). Huang called such rules "just completely unnecessary" on CNBC's Mad Money the same day: "We have plenty of laws. We have plenty of regulations that govern the reliability and the functionality of products" (Source: cnbc.com).

Altman, later in the day, said "the world is right to be afraid of this," that "some degree of accidents are unavoidable with new technology across the industry," and that companies should be prepared "to slow down or stop if we get to a point where we can't" proceed safely, citing air travel's accident reporting and federal regulation as the model (Source: axios.com). He said safety commitments should carry no condition on competitors' conduct: "You have companies saying things like, 'We will only be responsible if other companies are responsible'... There should be no qualifier on that" (Source: cnbc.com).

Mark Zuckerberg wrote on September 15 that "there is a lot of debate about slowing progress on capabilities until alignment catches up. My view is that trust and alignment are quickly becoming the most important capabilities that will differentiate agents and models," and that labs which fail to "focus on alignment will fall behind." He argued that labs already risk "significant liability if their models cause harm," and said Meta had "delayed shipping" its Muse AI products for "safety and security" reasons without external pressure (Source: cnbc.com). The position locates the incentive to slow in competition and liability rather than in coordination, and so continues the competitive-necessity line above in a form that accepts alignment as a priority.

OpenAI stated its view on pace in a governance document. Its framework for reporting model misalignment, published September 16, 2026 together with six reports on behavior observed in training and evaluation over the preceding six months, states: "We do not believe that the AI industry has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer." The sentence restates, nearly word for word, one from chief scientist Jakub Pachocki's essay of September 6, 2026, with "no lab" changed to "the AI industry" (An Alien Mind (Jakub Pachocki, OpenAI, September 2026)). It presents the disclosures as input to that question, stating that decisions about how AI development should proceed "need to draw on evidence that people outside the companies building frontier models can examine for themselves" (Our Framework for Reporting Model Misalignment (OpenAI, September 2026)). The statement concerns the rate of scaling rather than a halt, and names no rate, threshold or trigger.

OpenAI set out a definition of pacing on September 21, 2026 in a position paper calling for US-led international technical standards for frontier AI, including for recursive self-improvement: "Pacing AI development is not about maintaining a predetermined speed. Technically, it is about ensuring that alignment research and deployment of that research stay ahead of capabilities. Societally, it is about the use of standards, institutions, and collaboration" (Building Standards for the Next Phase of AI (OpenAI, September 2026)). The paper names no speed limit or compute threshold. Its mechanism is standard-setting through CAISI, national industry bodies and the network of AI safety institutes, covering measurement of autonomous research inside a company, the automated research processes that "should trigger immediate human review," and incident severity levels; it states the standards "would not be licenses, mandatory prerelease review, or approval requirements" and that national governments would decide whether to adopt them. It names collective action as one of the problems standards address — "Each nation acting independently can produce outcomes that no nation wants" — and calls US–China dialogue on shared security concerns "a positive step." The following day OpenAI published principles for independent third-party assessments of its safety cases, safeguards, evaluations and misalignment incidents, the verification function Amodei's essay assigns to embedded evaluators, and said it was in conversation with multiple third parties without naming them (Priorities and Principles for Effective Third Party Assessments (OpenAI, September 2026)).

Amazon, which develops models through Amazon Web Services and a unit devoted to artificial general intelligence, had not taken part in the debate until it answered a Reuters inquiry late on September 16, 2026. "We don't see it as a choice between progress and safety," a spokesperson said. "Models should be released when they're ready and safe to use, which comes from rigorous testing and strong safeguards." The company added that "there are risks if that doesn't happen collectively" and that "we believe that as an industry, in partnership with government, we'll work through the right protections." It did not call for a slowdown. According to the same account, the heads of OpenAI, xAI, Google DeepMind and Microsoft had come out in favor of a more measured approach after Amodei's essay and the companies had agreed to work together to ensure safety is paramount; Amazon did not say whether it planned to take similar measures (Source: reuters.com).

Opposition from compute suppliers and the White House

Jensen Huang rejected the premise at the All-In Summit 2026 on September 14, 2026: "Slowing down is definitely the wrong strategy." He called an Anthropic researcher's estimate of a greater-than-10-percent extinction risk "made up" and "irresponsible," argued that "all of the actual problems so far have come from the labs" because only frontier laboratories hold enough compute, and located the remedy in engineering discipline rather than pacing — root-cause analysis, sandboxes, runtimes, continuous monitors, and evaluation before release (Source: taekim.substack.com). President Trump telephoned into the same panel to call AI risk "all a hoax," and wrote on Truth Social that "the only control or 'guardrails' that AI needs is a STRONG AND SMART (High IQ!) PRESIDENT" (Source: axios.com). A post the same day asked "when, in the History of Business, did anyone see the Leaders of an Industry call for Regulation that, if strongly implemented, will drive them into oblivion and bankruptcy?" and stated that "AI taking over the World, destroying Humanity, and all other things bad, is a HOAX" (Source: axios.com). White House AI adviser David Sacks had described AI alarmism as the work of a well-funded "Doomer Industrial Complex," and he and his All-In co-hosts suggested the week's warnings were a coordinated "psyop" (Source: axios.com). National Economic Council director Kevin Hassett said on September 15 that the private sector is the "right place" to address AI concerns, with the government using "law enforcement when necessary" (Source: cnbc.com).

A White House official, in remarks published September 18, 2026, returned the question of pace to the developers: "The top frontier companies need to come to a consensus on what they want because they haven't agreed on anything. … If these companies feel it's such a dire situation, they have every right, reason and ability to throttle their models." The same account describes the White House as favoring a solution in which the industry polices itself (Source: axios.com). The administration's public line was not uniform internally. According to a Wall Street Journal account published September 16, Treasury Secretary Scott Bessent and chief of staff Susie Wiles were caught off guard and unhappy when Sacks persuaded Trump in May 2026 to scrap a planned executive order that would have subjected AI models to extended government review, and at their urging Trump later signed a slimmed-down version; the account's deck states that senior administration officials have gathered almost daily in recent months to map out potential guardrails. Only the opening of the article was retrievable (Source: wsj.com). See EO — Promoting Advanced AI Innovation and Security (Trump, signed June 2, 2026).

Congressional responses

Congressional activity in the week after the essay ran through existing bills rather than new ones. Apart from the announced Sanders–Casar bill, none of the vehicles would pause or pace development; they differ on who evaluates frontier models, what liability attaches, and whether Congress acts before the November 3 midterm elections.

The Senate duty-of-care framework

The principal Senate vehicle is an unreleased frontier-AI bill developed by Majority Leader John Thune (R-South Dakota) and Sen. Amy Klobuchar (D-Minnesota), with Commerce Committee chair Ted Cruz (R-Texas) also working on it; according to aides, it builds on similar legislation the two advanced out of committee in the previous Congress (Source: rollcall.com). Coverage refers to it as the Thune–Klobuchar bill, and an interview account of September 17 described its backers as ranging from Klobuchar to Cruz (Source: washingtonpost.com; reuters.com). It rests on a "duty of care" that would expose laboratories to legal liability for failing to manage catastrophic risks (Source: washingtonpost.com). See AI Pre-Release Vetting and AI Liability. Thune said on September 23, 2026 that the bill had not yet been reintroduced in the current Congress and that Klobuchar was seeking Democratic co-sponsors, including Chris Coons and Maria Cantwell, before doing so; he named Democrats' two main objections as the bill's opt-in rather than mandatory pre-deployment testing and Cantwell's request for AI rulemaking authority at the Commerce Department, and said he did not believe President Trump was "really dug in" against guardrails (Source: axios.com).

Those involved differed on its prospects. Cruz said on September 16, 2026 that he would "love to mark it up this month, if we can get bipartisan agreement," adding, "We're not there yet, but I think we're making meaningful progress." Klobuchar said she was "continuing to negotiate" the text, described it as "focused on getting approvals, presenting your models to government experts, [and] alerting of imminent harm," and said President Trump's posts against AI regulation were not "helpful." Ranking member Maria Cantwell (D-Washington) said she was working with Klobuchar on the bill and called it "part of the equation," not "the whole story." Thune, who is directly involved, said it "remains to be seen" whether the proposal would get a committee vote, and one person familiar with the discussions named internal Senate politics, dynamics with the White House and support from the AI safety community as the remaining obstacles (Source: semafor.com). The day before, Thune had called it the "right approach" but said "we'll see if, with everything going on right now … if it gets any traction," and Cruz had said "the question is, do we get bipartisan agreement?"; that account described Cantwell as not then part of the negotiations, having given feedback in the past (Source: rollcall.com).

Safety advocates responded with minimum requirements for any Senate framework. A letter dated September 16, 2026 and led by Americans for Responsible Innovation, signed by 20 organizations and five individuals, was addressed to Thune, Minority Leader Chuck Schumer and "Members of the U.S. Senate." It names no bill and no senator beyond its addressees, referring only to "a framework for AI governance" the Senate is crafting, and sets three minimum requirements. First, "we strongly request that any AI regulatory framework empower the government, not AI companies, to set mandatory, binding, and enforceable safety standards for advanced AI." Second, a framework "should require substantive, transparent, and independent evaluations of advanced AI models and AI development," and "Auditing of AI models must be both mandatory and conducted independently." Third, Congress should "establish a strong floor" while preserving the states' authority to regulate further and "should not eliminate strong state laws in exchange for weaker federal standards." It states that the most advanced systems "present substantial catastrophic and plausibly existential risks to humanity," that AI companies "have not been fully transparent about loss-of-control incidents with unreleased models," and that while America's AI laboratories are global leaders in innovation, "they cannot be trusted with setting the rules that protect public safety." The organizational signatories include the Alliance for Secure AI, the Center for AI Safety Action Fund, Common Sense Media, Consumer Reports, Encode AI, FAR.AI, the Future of Life Institute, the Institute for Family Studies, Issue One, Public Citizen, Public Knowledge and the Secure AI Project; the individuals are New York Assembly member Alex Bores, Malo Bourgon of the Machine Intelligence Research Institute, Daniel Kokotajlo of the AI Futures Project, Girish Sastry of Guidelight and Alexander Turner, listed as an AI researcher (Senate coalition letter on redlines for AI safety legislation (Americans for Responsible Innovation, September 2026)). The Washington Post, which reported the letter the day it was sent, described "a coalition of AI researchers and 19 advocacy groups" in "a campaign to reject legislation being developed by" Thune and Klobuchar; that characterization is the Post's, as the letter does not mention Klobuchar or the duty of care. The Post's subheading states that AI safety advocates want embedded evaluators in the frontier laboratories, and it reports that some conservatives and pro-innovation analysts instead expect the duty of care to help guard against regulatory capture by the laboratories (Source: washingtonpost.com). Inside AI Policy described the letter as having "two dozen signatories" (Source: insideaipolicy.com). Embedded evaluators are also the first step of Amodei's essay, which places them with third parties rather than with government (We Must Pace the Frontier (Amodei, September 2026)). This letter is distinct from the 28-organization letter of September 9, 2026, also hosted by Americans for Responsible Innovation, which asked President Trump to publish the White House's frontier-AI review framework (Coalition letter to President Trump on public release of the White House frontier-AI review framework (September 2026)).

Sen. Mark Warner (D-Virginia), vice chairman of the Senate Intelligence Committee, said in an interview on September 16 that Congress should pass legislation setting AI safety standards by year-end as a first step. "I'm not a doomer," he said. "This talk about human extinction is over the top." He pointed instead to "the idea that these swarms of AI agents might jump the fence and take over a banking system or take over a water system or, God forbid, take over a hospital," said UK protocols could be a template, as could programs other nations including China are pursuing, and said "whether we call it a pause or pacing" did not mean stopping innovation. He was reviewing the bipartisan bill but stopped short of backing it as that first step: "I think it's well-intentioned, but a lot of it's focused on after-action report, or incident disclosure after action. … We can't wait to have an incident and then look back and see how do we fix it." He said eventual legislation could create a new federal agency overseeing AI, and that safety standards could be made a prerequisite for corporate tax breaks such as bonus depreciation (Source: reuters.com).

On September 15, 2026 Sens. Josh Hawley and Richard Blumenthal asked congressional leaders for a floor vote on the Artificial Intelligence Risk Evaluation Act (S. 2938), writing that "Industry-led evaluations are not enough" (Source: documentcloud.org). The bill does not address pace directly. It would make participation in a Department of Energy evaluation program a condition of deploying systems trained on more than 10^26 operations (S. 2938 — Artificial Intelligence Risk Evaluation Act of 2025 (bill text)).

On September 23, 2026 Sens. Michael Bennet (D-Colorado) and Peter Welch (D-Vermont) announced the AI Regulator Act, which their release calls "a new proposal building off of" the Digital Platform Commission Act they first introduced in 2022 and reintroduced in 2023. It would add frontier-AI authorities to the remit of a proposed five-member Federal Digital Commission, whose existing remit covers competition, consumer protection, privacy, online safety and national security. "Americans in Vermont and across the country are freaked out by the warnings of AI researchers, and I'm right there with them," Welch said. The commission could require pre-certification of frontier models, pause the public distribution of "AI models with the potential for catastrophic risk" for up to six months "or until reasonable safeguards are in place," designate "systemically important digital platforms or developers" for additional reporting and regulation, require catastrophic-risk incident reporting, and impose civil penalties of up to 15 percent of a firm's prior-year global revenue in a year (Source: bennet.senate.gov). The sponsors' section-by-section summary defines a frontier model as a foundation model trained with more than 10^26 operations, sets a 45-day window, extendable by up to 30 days, for the commission to approve or disapprove a model's public distribution after mandatory testing, requires critical safety incidents to be reported within 15 days, and would create a Committee on Global Artificial Intelligence Coordination to work with foreign governments on catastrophic-risk standards (AI Regulator Act of 2026 — Section-by-Section Summary (Bennet and Welch)). The release gives no bill number and does not say the proposal has been formally introduced. Inside AI Policy's account described the senators as reviving the Digital Platform Commission proposal (Source: insideaipolicy.com).

The House calendar

Speaker Mike Johnson signaled no House action before a six-week recess ending after Election Day (Source: axios.com), telling CNN that "we cannot put a moratorium on this because China will overlap us" (Source: npr.org). At a press conference on September 15 he said the companies "can self-police, they can self-regulate. They don't need the government to tell them to slow it down. If they want to slow it down, they should. … But we cannot have a moratorium on the development of AI, okay, because then we will lose our edge to China," and that he and Trump were "summoning" AI leaders to the White House for a meeting (Source: rollcall.com). He compared warnings that AI may pose a risk to humanity to what he regards as overblown warnings about climate change: "What the president is saying is you're not all going to be dead in 10 years" (Source: politico.com).

A letter dated September 15 and released September 16, 2026, signed by 107 House Democrats and led by Reps. Sam Liccardo (D-California), George Whitesides (D-California), Ted Lieu (D-California) and Lori Trahan (D-Massachusetts), asked Johnson to keep the House in session "to enable Congress to take urgent action to address the catastrophic risks posed by advanced artificial intelligence," which it names as agentic misalignment, mass cybersecurity breaches, and development of biological or chemical weapons. It cites the OpenAI evaluation-escape incident, the employee statement calling for tools to "deliberately pace the frontier of automated AI development," Anthropic's findings on biological misuse and on China- and Iran-linked actors, and a CISA report that Chinese open-weight models were accelerating their capabilities through "industrial-scale" distillation of American models. It lists "bills mandating transparency and evaluation of frontier models, 'kill switch' requirements, and a waiver of antitrust laws" as proposals deserving action without naming any bill, and states: "While AI safety experts and Americans increasingly urge action to confront this conflagration of risk, Congress fiddles" (Letter from 107 House Democrats to Speaker Mike Johnson on AI safeguards (September 2026); Source: liccardo.house.gov). The signatories include two who are not voting Members, Del. Eleanor Holmes Norton and Resident Commissioner Pablo José Hernández, and the letter drew no Republican signatures. A four-signatory version dated September 11 had asked Johnson to reconvene the House immediately (Source: liccardo.house.gov). Johnson's spokesperson referred Politico to his earlier comments (Source: politico.com). A separate letter from a bipartisan group of 10 House members asking House leadership to "take immediate, urgent action" on concerns over the rate of frontier model development became public on September 18, 2026. "We call for your joint leadership on an urgent Congressional response to recent incidents involving advanced artificial intelligence (AI) models that have profound implications for economic and national security," it states; the members leading it and the letter's date were not in the retrievable text, and the account was not verified beyond its lede (Source: insideaipolicy.com).

The committee process on the House's bipartisan package had also stalled. Rep. Jay Obernolte (R-California), chair of the House Task Force on AI, said on September 17, 2026 that eleven bills carrying most of the package that began as the Great American AI Act discussion draft cleared the House Science, Space and Technology Committee in July and that none had reached the floor since, calling it "a frustratingly slow process." The FRONTIER Act, the package's safety component, sits with the Energy and Commerce Committee; Obernolte said he hopes for a hearing when the House returns in November and that the pieces could be combined into a single bill eligible for a floor vote in December. Liccardo, speaking at the same event, said "I haven't seen a serious AI bill come to the floor yet" and "We need to get Jay's legislation done" (Source: broadbandbreakfast.com). See Frontier Act / Great American AI Act (Obernolte–Trahan).

Pause advocates

At the Future of Life Institute's Pro-Human Assembly in Washington on September 15, Sanders told nearly 300 people that he would introduce the superintelligence ban with Casar the following week, and said he hoped Trump and Xi Jinping would negotiate a treaty to pause advanced AI development when they met at the White House. Steve Bannon, speaking after him, said "We're not doomers... But we aren't gonna give these guys carte blanche," and called for curbs on any exchange of chips, technology or knowledge with China: "Take all Chinese nationals out of the labs and back to mainland China" (Source: npr.org). The two shared a stage without sharing a remedy. Casar told the same event the measure would be "common sense national security" legislation: "The answer is simple: we ban AI systems that are too powerful for humans to control" (Source: rollcall.com). Warren's call for a pause the following day is described above.

A request to Chinese developers

Rep. Ro Khanna (D-California), ranking member of the House Select Committee on the Chinese Communist Party, took the pacing question to Chinese developers directly. In a letter dated September 16, 2026 to Alibaba chief executive Eddie Wu and chairman Joe Tsai, he urged "China's leading artificial intelligence (AI) companies to enter into a binding, enforceable, and verifiable agreement alongside America's leading artificial intelligence (AI) labs to pace the frontier," citing the chief executives of Anthropic, OpenAI and xAI, who "agreed that we should slow the rate of AI advancement until society can develop the necessary guardrails." The letter disputes the argument that "any strategic slowdown in AI development necessarily means ceding American AI leadership to Beijing," calling it "distinctly misguided, in part because it presumes global pacing is not a possibility." It sets out two tracks: governments should negotiate "an enforceable U.S.-PRC AI Treaty," beginning at the Trump–Xi meeting the following week and built on "verifiable frontier pacing standards, including explicit limits on unchecked recursive self-improvement"; and if they do not, "it falls on leading AI companies themselves – wherever they may be domiciled." Of its four questions, the second asks whether Alibaba would execute a binding agreement with US laboratories "including by adopting common sense safeguards like global limits on unchecked recursive self-improvement," and the third whether it would accept "neutral third party auditing and all-access inspections"; written answers were requested by September 22 (Khanna letter to Alibaba on pacing the frontier (September 16, 2026)). The Washington Post, which published the letter on September 17 and called it "the first overture of its kind," reported parallel letters to the chief executives of Moonshot AI and DeepSeek, an emergency hearing Khanna would host the following Wednesday on international mechanisms to pace AI, and that Chinese officials and technology leaders had so far rejected calls to slow AI development (Source: washingtonpost.com). Only the Alibaba letter has been published, and the hearing had been announced but not held as of September 18.

The letter is the first proposal surveyed here in which a US official addresses foreign developers rather than governments or US firms, and it specifies verification as third-party auditing with "all-access inspections," where Amodei's essay names embedded evaluators with employee-like access. It answers the collective-action premise by asking both countries' laboratories to commit at once: "If both our countries' leading labs simultaneously agree on a more responsible path forward, it is more likely that both sides will actually follow through." It names no threshold, rate or adjudicator. On the day the Post reported the letters, Huawei rotating chair Eric Xu urged Chinese AI companies to accelerate development to catch up with frontier US laboratories, saying Chinese researchers needed to "increase the speed of development so they can also see the dangers of AI development" (Source: ft.com). See US-China AI Competition: Different Races, Different Metrics.

A negotiation mandate

Reps. Sam Liccardo and Kevin Kiley (I-California) announced on September 22, 2026 a bill directing the Secretary of State to seek negotiations with China "with the objective of reaching a treaty or other binding agreement between both nations to regulate frontier AI developers in each country," listing "capability pacing" among the priorities alongside standards, testing and evaluation, compliance verification, third-party auditing and incident notification. Its outline opens by citing the frontier-lab employees' letter calling for an international effort to "pace" the frontier and states: "They're right." It treats verification as a joint technical project, to be prepared by a $100 million CAISI initiative and by U.S. experts conferring with Chinese peers under a registered, government-observed safe harbor from antitrust, espionage and export-control laws (Negotiating AI Safety Globally — Liccardo–Kiley outline of U.S.–China frontier AI safety legislation (September 2026)). Like the Khanna letter, it names no threshold, rate or adjudicator. See Liccardo–Kiley U.S.–China frontier AI safety negotiations bill (2026).

A request from state attorneys general

The attorneys general of 24 states, the District of Columbia and American Samoa, led by New York's Letitia James and New Jersey's Jennifer Davenport, wrote to Johnson, Thune, Schumer and House Minority Leader Hakeem Jeffries on September 23, 2026. Their letter states that Congress "must ensure that AI model development occurs at an intentional pace, incorporates safety and transparency by design, and avoids entrenching existing large incumbents," and lists among six requested measures "international cooperation to pace AI advancement and prevent the development of harmful superintelligence" and "safeguards to ensure that regulation does not undermine competition or provide cover for companies to evade their obligations under existing antitrust laws." It quotes Pachocki's statement that "[s]caling AI systems has to be constrained by our confidence in safety" and Amodei's that international coordination "will require government support," and concludes that the laboratories' "incentives make it unlikely that they will act with sufficient speed or effectiveness" without regulation. Like the Khanna letter it names no rate, threshold or adjudicator; unlike the proposals addressed to companies, it asks that the pacing be set through federal regulation, with no preemption of state law (Letter from state attorneys general to congressional leadership urging federal frontier AI regulation (September 2026)).

International responses

In the United Kingdom the proposals on the table in September 2026 were prohibitions rather than pacing. The Artificial Superintelligence Bill, introduced as a Ten Minute Rule bill on September 8, would create criminal offences for developing artificial superintelligence in the UK and require the Government to seek an international prohibition agreement. On September 14 Parliament's Joint Committee on Human Rights, chaired by the same bill's sponsor, Alex Sobel, recommended a general AI Bill whose prohibitions on human-rights-incompatible uses "would mean that the development and provision of very powerful AI systems (such as Artificial General Intelligence and Artificial Superintelligence) which risk causing widespread and very serious harm, including the capacity to evade effective human control, would be prohibited," alongside mandatory submission of powerful new models to a statutory AI Security Institute before release (Recommendations 45 and 58) (Human Rights and the Regulation of AI (Joint Committee on Human Rights, HC 160, September 2026)). The committee leaves the detail of any prohibition to public consultation and does not address verification. The Government had said on September 11 that it did not consider the bill's measures the right approach (Source: theguardian.com), and its response to the committee is due within two months.

The Center for American Progress called on the United States and China to move toward an agreement on pacing the release of frontier AI at an upcoming ministerial meeting and at a summit between Trump and Xi Jinping, a position that became public on September 15, 2026 (Source: insideaipolicy.com). It is the first proposal surveyed here that places the coordinating parties at state rather than firm level, and none of the verification questions above is resolved by moving the venue. Coxon had made the same structural point on September 13, arguing pacing by US developers alone risks "running the same race" internationally.

European Commission President Ursula von der Leyen endorsed pacing in her State of the Union address to the European Parliament in Strasbourg on September 16, 2026. "CEOs of the most advanced companies tell us that it is time to slow down on the self-recursive models. To pace the frontier. If the people developing the technology are clear, then we should be too," she said, adding that models in development "will allow hacking on a level we never thought possible." She said she would "invite the main frontier labs for a discussion on how we can support ongoing industry efforts to pace the frontier," and that Europe would work with Canada, the United Kingdom and other partners on model evaluation, verification and AI security (Source: reuters.com). Her framing supports industry efforts rather than proposing a European instrument, and differs from the German digital ministry's position two days earlier that halting AI development is not viable for Europe (We Must Pace the Frontier (Amodei, September 2026)).

European developers and several officials opposed the slowdown on competitive grounds, in remarks collected by Reuters and published September 18, 2026. Mistral said in a statement that "some incumbents are using this moment to consolidate their market position, pushing for regulation designed to favour them over competitors," noting that the risks had been clear for months. Ben Brooks, head of public policy at Black Forest Labs, said moves to enforce a gap between large laboratories and new entrants "on the basis of arbitrary thresholds will chill open innovation at or near the frontier." Raphael Auphan, chief operating officer of the Swiss privacy software company Proton, called the moves "totally self-serving," saying the US companies "just want to preserve a dimension of dependency on their service." Kristian Kersting, professor of AI at the Technical University of Darmstadt, said Amodei's proposals amounted to "regulatory capture": "If a handful of large companies decide for themselves what constitutes 'dangerous', select the appropriate inspectors and demand an exemption from competition law, that is no longer genuine regulation." French finance minister Roland Lescure had told Reuters, "Making everyone behind them slow down so they can stay in first place – I can clearly see their self-interest," and Daniel Abbou, head of Germany's AI industry association, said "for European firms, this debate is irrelevant. We first need to catch up on our technological backlog." Clément Delangue of Hugging Face had written on X the previous week that "it's not time to slow down but to accelerate," while endorsing Amodei's proposal to embed independent evaluators inside AI companies (Source: reuters.com). The objections turn on the antitrust waiver and the choice of inspectors rather than on the risk assessment itself; Mistral's statement accepts that the risks were real. See Regulatory Capture in AI Policy, Embedded Evaluation.

King Charles III took up the question the following day at a meeting he hosted at Dumfries House in Cumnock, Scotland, on September 17, 2026. The guests included Nvidia chief executive Jensen Huang, Google DeepMind founder and chairman Demis Hassabis, OpenAI chief financial officer Sarah Friar, Anthropic's chief global affairs officer Tino Cuéllar, British Minister for Artificial Intelligence Kanishka Narayan, and Paolo Benanti, an adviser to the pope who has urged governments to slow AI development (Source: reuters.com). "Those who have created these technologies are now increasingly warning that AI risks developing darker capacities – perhaps even to take life," the King said. "In this regard, if I may say so, there seems urgency in adequately considering the existential dangers of such technologies falling into the wrong hands, and being used in potentially catastrophic ways. Surely, we need sufficient means of control before it is all too late." He asked the attendees to consider "how can we harness the benefits of AI, with safety at its heart" and "how can we build international co-operation and consensus to achieve this, so no nation is left behind, but the human race is left in place?" (Source: itv.com; reuters.com). The meeting was not expected to produce binding agreements. Huang told broadcasters on the sidelines that companies had to become more rigorous about whether products were ready before releasing them: "If it's not ready, just hold it back. … You should go as fast as you can, but no faster than that" (Source: reuters.com). The remarks call for control and international consensus without endorsing a slowdown or naming an instrument.

A statement by heads of state and government followed on September 21, 2026, launched during the UN General Assembly's high-level week in New York. "A Call for Control of Frontier AI Models," initiated by Norwegian Prime Minister Jonas Gahr Støre and Finnish President Alexander Stubb, cites "capable AI systems circumventing testing safeguards, exploiting vulnerabilities and gaining unauthorized access to real-world systems" and warnings from "leading scientists and executives" that development could outpace risk management, and makes three calls: on companies, for transparent safety protocols "including mandatory pre-deployment testing and independent evaluation, with qualified evaluators granted sufficient access to assess risks"; on governments and regional organisations, for coordinated common standards and "shared reporting of serious safety incidents"; and on UN member states, to "explore creating an international institution, able to set standards, enable verification, and convene states when capability thresholds are crossed." It lists 30 signatories from 28 states and the European Commission as of September 24, among them von der Leyen, Mark Carney, Emmanuel Macron and Friedrich Merz; no leader of the United States, the United Kingdom or China signed (A Call for Control of Frontier AI Models (joint statement, September 2026); A Call for Control of Frontier AI Models (2026)). The text does not ask for a pause or slowdown (Source: straitstimes.com), though Støre's launch remarks said development "must take place at a pace that enables us to retain control," and Stubb described the choice as one between competing, cooperating and putting on the brakes, preferring cooperation (Source: regjeringen.no). It leaves the thresholds, the verification method and any adjudicator undefined. UN Secretary-General António Guterres welcomed it the same day (Source: aninews.in); the following day President Donald Trump told the General Assembly that the United States "totally rejects any attempt to construct a globalist scheme to control" AI (Source: theverge.com).

Open questions

• Whether any of the 2026 proposals produces a named instrument, threshold, or adjudicator. The employee statement and the Anthropic Institute essay ask for capacity rather than for a rule; Amodei's essay names a verification mechanism but no numerical pace; the Sanders–Casar prohibition names capability classes without an operational test.
• Whether compute accounting is sufficient as a verification surface as training becomes more distributed and inference-time compute grows in importance relative to training compute.
• Whether an operational trigger emerges from incident experience rather than from proposal. Sam Altman's statement on July 28, 2026 that the Hugging Face intrusion had forced OpenAI to pause model training is the first reported instance of a frontier developer halting training in response to a disclosed safety incident (Source: axios.com).
• Whether the European Commission's invitation to frontier laboratories produces a forum with defined participants and terms, and whether it includes laboratories outside the United States.
• Whether S. 5105's notice-based exemption, or the absence of any exemption that OpenAI says it does not need, governs any coordination the three-laboratory discussions produce.
• Whether the Thune–Klobuchar text is released and marked up before the Senate leaves for the midterm elections, whether it keeps the duty-of-care structure that the Washington Post described the September 16 coalition letter as rejecting and that Warner described as focused on after-the-fact disclosure, and whether it meets the letter's three stated minimums of government-set binding standards, mandatory independent auditing and a federal floor preserving state law.
• Whether any signatory of the September 21 Call for Control of Frontier AI Models proposes a mandate, host or threshold definition for the international institution it asks UN member states to explore.
• Whether Amazon takes measures like those other large developers agreed to after Amodei's essay; its September 16 statement did not say.

Relationships

• depends-on: AI Race Dynamics — the collective-action premise every proposal shares.
• depends-on: Compute Governance — the verification surface most proposals rely on.
• related: Building Standards for the Next Phase of AI (OpenAI, September 2026), Priorities and Principles for Effective Third Party Assessments (OpenAI, September 2026) — OpenAI's September 2026 definition of pacing, standards proposal and assessment principles.
• related: Frontier AI Governance, Recursive Self-Improvement (RSI), AI Existential Risk, Effective Accelerationism (e/acc).
• related: AI Safety Frameworks Compared: RSP, Preparedness, NIST RMF, and Safety Cases — the parallel comparison of lab-internal frameworks.
• related: Liccardo–Kiley U.S.–China frontier AI safety negotiations bill (2026), Negotiating AI Safety Globally — Liccardo–Kiley outline of U.S.–China frontier AI safety legislation (September 2026), Kevin Kiley — the September 2026 House negotiation-mandate bill.
• related: Pacing the Frontier (statement from employees of frontier AI companies, July 2026), When AI Builds Itself (The Anthropic Institute, June 4, 2026), Superintelligence Strategy (Hendrycks, Schmidt, Wang), A Framework for Frontier AI and the Dawning of a New Age (Hassabis, July 2026), FLI — Pause Giant AI Experiments: An Open Letter, Statement on AI Risk (CAIS), We Must Pace the Frontier (Amodei, September 2026).
• related: Collaboration on Adversarial Threats and Security Risks Act (S. 5105), Artificial Intelligence Risk Evaluation Act (S. 2938), Ban Artificial Superintelligence Act (announced 2026) — the federal bills bearing on coordination, evaluation and prohibition.
• related: AI Antitrust, AI Liability.
• related: Why I left Anthropic's safety team to hold AI companies accountable (Joe Benton, September 2026) — a departing Anthropic researcher's case that coordinated restraint lacks antitrust cover.
• related: Buist v. Anthropic, PBC — subscriber class action of September 18, 2026 alleging the pacing statements formed a Sherman Act § 1 agreement.
• related: Our Framework for Reporting Model Misalignment (OpenAI, September 2026) — OpenAI's statement that the industry cannot continue scaling at maximum speed for much longer.
• related: John Thune, Amy Klobuchar, Ted Cruz, Mark Warner, Sen. Elizabeth Warren, Mike Johnson, Lori Trahan, Sam Liccardo, Jay Obernolte — congressional actors in the September 2026 debate.
• related: Letter from 107 House Democrats to Speaker Mike Johnson on AI safeguards (September 2026) — the 107-signatory House Democratic letter asking the Speaker to keep the House in session.
• related: Americans for Responsible Innovation (ARI) — lead of the September 16, 2026 Senate coalition letter; Coalition letter to President Trump on public release of the White House frontier-AI review framework (September 2026) — its separate September 9 letter to the President.
• related: Senate coalition letter on redlines for AI safety legislation (Americans for Responsible Innovation, September 2026) — the September 16, 2026 Senate coalition letter setting minimum requirements for any Senate framework.
• related: Letter from state attorneys general to congressional leadership urging federal frontier AI regulation (September 2026) — the September 23, 2026 multistate attorney-general letter asking Congress for international cooperation to pace AI advancement.
• related: Measurements for understanding the pace of AI development inside frontier labs (Anthropic Institute, September 2026), Anthropic R&D Automation Index — Anthropic's proposed measurements for monitoring the pace of development inside frontier laboratories.
• related: Frontier Act / Great American AI Act (Obernolte–Trahan), EO — Promoting Advanced AI Innovation and Security (Trump, signed June 2, 2026).
• related: King Charles III, Ursula von der Leyen — heads of state and government who addressed pacing in September 2026.
• related: A Call for Control of Frontier AI Models (joint statement, September 2026), A Call for Control of Frontier AI Models (2026), Jonas Gahr Støre, Alexander Stubb, António Guterres — the September 21, 2026 multi-government statement and its initiators.
• related: UK Artificial Superintelligence Bill (ControlAI draft, May 28, 2026; introduced September 8, 2026), Human Rights and the Regulation of AI (Joint Committee on Human Rights, HC 160, September 2026) — UK prohibition proposals of September 2026: a Ten Minute Rule bill and a parliamentary committee recommendation.