The Moral Race: How Rhetoric Became America's AI Alliance Architecture

On May 4, Kevin Hassett compared the government's new pre-deployment AI review framework to "FDA drug approvals" ^[1]. The analogy was domestic, regulatory, cautious. A volunteer testing body called CAISI was described as advancing "the state of AI security" ^[1]. Even JD Vance, pushing hard, framed his objection in safety's vocabulary: he said the AI future would not be won by "hand-wringing about safety" ^[1]. The argument was about how much safety, not whether safety was the right frame. Seven weeks later, on June 25, Representative Jim Banks called AI competition with China "a moral race for our country" ^[2]. His colleague Brian Mast cast America as "the superhero" in a superhero-versus-supervillain contest with Beijing ^[2]. Treasury Secretary Scott Bessent named China surpassing the US in AI as a primary national risk ^[2]. The vocabulary had moved from pharmacology to civilizational stakes. Something happened between those two dates that changed the way American officials talk about AI control in public, and the change produced an institution. That institution is Pax Silica. On June 25 and 26, the State Department hosted 35 nations to sign the Joint Statement on AI Opportunity, led by Under Secretary of State Jacob Helberg. New signatories included the EU, Germany, the Netherlands, Greece, Kazakhstan, Panama, Argentina, Chile, and Costa Rica. The goal is to secure the full AI supply chain: semiconductors, critical minerals, energy, the works ^[3]. Helberg's framing was explicit:

The future of AI will not be determined by who regulates first. It will be determined by who builds first and builds the most capacity. More energy. More compute. More chips. More talent. More builders. A future where we accelerate innovation and private investment is welcomed — Jacob Helberg

. David Landau, a senior State Department official, called the allies' supply-chain interdependence "our strength, not our weakness" ^[3]. This is not a safety framework. It is a capacity-building bloc. To see what Pax Silica is and why it exists in the form it does, you have to trace the arc from that FDA analogy to that moral race, because the arc is the mechanism. The toggle came first. The Commerce Department issued export-control directives that forced Anthropic to disable its Fable 5 and Mythos 5 models for all foreign nationals ^[4]. The cutoff hit TCS, Infosys, and Indian startups directly. India's response was sovereign AI: NITI Aayog was tasked with reducing strategic dependence on US-based firms, and the IndiaAI Mission expanded to fund local data centers and GPU clusters. A founder at Sarvam AI put the lesson plainly:

The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. — Anthropic

^[4]. India was not alone. At the G7 summit in mid-June, France's Sébastien Lecornu declared that "we cannot rely on tools developed by foreign powers" and that "France must have its own tools" ^[5]. After the Five Eyes cyber chiefs warned on June 22 that AI threats were "months, not years" away, Canada's Mark Carney used the occasion to call for diversification away from US-controlled models: the overreliance on Mythos and Fable, he said, was something "we will have done something wrong" if they accepted without building alternatives ^[6]. The toggle's coercion produced exactly what coercion produces among sovereign states: a demand for autonomy. France, India, and Canada each said, in different registers, that they would not accept dependency on tools a foreign power could switch off. While allies were pushing back, the bilateral safety track with China was collapsing. On May 14, Trump and Xi met in Beijing and discussed "standard guardrails" for AI. The summit concluded without a signed governance framework or a formal agreement on AI risk dialogue ^[7]. Bessent's language in the aftermath was transitional. He still used the word "safety" but wrapped it inside a competitive calculus, saying the US could lead safety discussions "because it maintains a technological advantage over China" and framing the goal as "the highest performance calculus where we can get the most innovation and the highest level of safety" ^[7]. Safety was no longer an absolute. It was one variable in an optimization problem the US intended to win. With the China safety track dead and allies chafing under the toggle, the race register consolidated. On June 2, Trump signed an executive order and a National Security Presidential Memorandum that integrated AI into the warfighting enterprise: a voluntary 30-day pre-release vetting framework for frontier models, with NSA classified benchmarking and a Treasury cybersecurity clearinghouse, plus a mandate for military AI acceleration and an autonomous weapons policy update within 90 days ^[8]. The language left no ambiguity about which register now governed:

My Administration will ensure that those who safeguard America and the American way of life are equipped with the most sophisticated and secure AI technologies to perform complex, time-sensitive, and highly-consequential missions, with full confidence that those tools will be available when they matter most … Through these efforts, my Administration will secure a decisive and enduring AI advantage against any and all adversaries while safeguarding the constitutional chain of command. — Donald Trump

By June 17, the G7 summit became the hinge where allied backlash met the race framing and was redirected. The same sovereignty impulses that Lecornu voiced could have produced fragmentation. Instead, AI executives at the summit, including Dario Amodei, Sam Altman, and Demis Hassabis, proposed a US-led global coalition and urged allies to "resist the temptation to splinter" into separate geopolitical blocs ^[5]. Macron proposed a "trusted partners framework" at the G7 ^[5]. The framework that emerged as Pax Silica carried the same logic: trusted partners get access stability in exchange for supply-chain alignment. India's MEIT Secretary sought and received assurances that frontier AI access would "remain stable for trusted partners" ^[3]. Material incentives drove the bargain. But the race framing supplied the political cover that made voluntary alignment defensible at home rather than subservient. It converted a dependency into a shared civilizational project.

2026-05-04 Hassett frames AI review as "FDA drug approvals"; safety register dominant ^[1]

2026-05-14 Beijing summit ends with no AI guardrails agreement; bilateral safety track with China fails ^[7]

2026-06-02 Trump executive order: "decisive and enduring AI advantage against any and all adversaries"; military register formalized ^[8]

2026-06-17 G7 summit: allied sovereignty demands meet race framing; Amodei urges allies not to splinter; Macron proposes trusted-partners framework ^[5]

2026-06-22 Five Eyes cyber chiefs warn AI threats are "months, not years" away; Carney calls for diversification from US models ^[6]

2026-06-25 Pax Silica summit: 35 nations sign Joint Statement; Banks calls AI competition "a moral race for our country" ^[3][2]

The framing is not coming from one place. Officials articulate it at the policy level ^[2]. Anthropic, the company most enmeshed in the control architecture, published a policy paper urging the US to "lock in a 12-to-24-month lead in frontier capabilities by 2028" and warning that Communist Party AI leadership would enable "repression at a scale that humans alone could not achieve" ^[9]. A nonprofit called Build American AI, linked to a tech-backed political fundraising network, paid social media influencers to frame Chinese AI progress as "a direct threat to U.S. national security, data privacy, and employment" through a two-phase campaign: positive US AI mentions in lifestyle content first, then explicit warnings about Chinese AI ^[10]. The evidence shows these actors promoting the same framing in parallel. It does not show coordination among them. What is clear is that by late June the race register is audible from the White House, from Congress, from a major AI lab, and from paid influencer content aimed at the public. The safety register has not vanished. It survives as the domestic enforcement layer. The pre-deployment review framework that Hassett compared to FDA drug approvals is still operating, and it is going international on its own track: New Zealand's NCSC gained access to Mythos AI under a cyber defense mandate, extending the testing gate to allied cybersecurity agencies ^[11]. Anthropic itself proposed a "global pause" over recursive self-improvement risks on June 4, with Jack Clark warning that "the AI industry has a gas pedal, but it doesn't have a brake pedal" ^[12]. OpenAI disagreed, arguing that "democratic governments, not private labs, must determine the rules" ^[12]. The safety register still produces policy proposals. But it operates at the industry and domestic level. The race register operates at the state and multilateral level. They are layered, not merged. And when the two registers conflict, the race register wins. Trump canceled a planned AI executive order specifically to ensure the US does not hinder its technological lead over China ^[13]. Hegseth said he "would not hesitate to reject AI models that won't allow you to fight wars" ^[13]. The DoD diversified to eight vendors after a single-model dependency scare, with Emil Michael stating the department would "never again be single-threaded with any one model" ^[14]. Inside the military, the safety voice still speaks. Admiral Bradley of US Special Operations Command cautioned at a special forces conference that humans must have confidence AI "is going to deliver violence only where we intend it to be delivered" ^[13]. But that voice is the marginal one. The race register is hegemonic at the political level. It does not merely coexist with safety; it overrides safety when the two point in different directions. The asymmetry with China is what makes the rhetorical overlay necessary. China is building mirror-image controls. A State Council decree effective July 1 forces unwinding of overseas AI transactions, bans unauthorized talent transfers, restricts travel for senior AI professionals, requires sign-off before accepting American capital, and authorizes retaliatory countermeasures ^[15]. Beijing blocked Meta's $2 billion Manus acquisition and elevated the decision to Xi's National Security Commission; it barred Manus co-founders from leaving the country ^[16]. These controls run on a pure sovereignty and national-security register. There is no "moral race" in China's framing, and there is no multilateral coalition. China does not need either, because it can command its firms directly. The US cannot command France, India, or Canada. It can cut off their access with the toggle, but that produces the backlash it got. It can offer them a seat at the gate through joint testing, but that does not secure a supply chain. Pax Silica is the instrument that does both, and the moral-race framing is the political glue that makes it defensible. The coalition runs on voluntary participation, not legal obligation ^[3]. There is no treaty. There is a Joint Statement. Material incentives, access-stability guarantees, and supply-chain alignment benefits hold the coalition together at the transactional level ^[3][4]. The race framing provides the ideological cover that makes voluntary alignment politically sustainable at home for each signatory. A government that joins a US-led supply-chain bloc to reduce dependence on China needs to tell its public a story about why. "We are managing AI safety risks" does not justify joining a geopolitical bloc. "We are winning a moral race against authoritarian AI" does. That is also the vulnerability. The sovereignty impulses the race framing manages have not disappeared. France still wants its own tools. India still distinguishes access from ownership. Canada still calls for diversification. These are not resolved by Pax Silica. They are absorbed into it, channeled by the framing into coalition participation rather than independent development. The framing holds as long as the China threat feels urgent and the US looks like the right partner to meet it. If either condition weakens, the political cover thins, and what remains is a voluntary coalition with no legal floor, held together by access guarantees that a future toggle could revoke. The architecture's most ambitious instrument is also the one with the shallowest foundation. The gate has regulation. The toggle has law. Pax Silica has rhetoric and incentives, and the rhetoric is what makes the incentives politically legible.