On 23 January 2025, President Trump signed Executive Order 14179, “Removing Barriers to American Leadership in Artificial Intelligence”. The order rescinds much of Joe Biden’s 2023 AI‑safety framework and hands primary oversight to the Commerce Department. A Commerce fact‑sheet hails the move as a decisive step toward global dominance; chip‑makers cheer looser guard‑rails. Yet the same stroke of the presidential pen also junked reporting duties meant to catch bias, safety failures and misuse before they spiral.
That volatility is the first problem with betting the world’s future on a single hegemon. American institutions can self‑correct—Congressional hearings, courts, whistle‑blowers—yet none stopped this pivot from “safety first” to “full speed ahead.” If the rules of engagement can change every four years, how can allies—or rivals—plan long term?
The second problem is strategic. The White House now frames AI as a zero‑sum duel with Beijing: keep cutting‑edge chips at home, slow China down, and U.S. models will stay on top. It sounds compelling until you see what scarcity does. In February, Beijing approved eleven new generative‑AI services, many trained on the last‑generation A800 graphics‑processing units (GPUs) Washington no longer blocks, according to Reuters. Chinese labs have doubled down on algorithmic efficiency—an arena tariffs cannot throttle.
Meanwhile U.S. export rules ricochet. Semiconductor groups warn that ambiguous thresholds and the threat of hefty fines for mis‑shipped silicon already freeze legitimate sales (Reuters, 12 March). On , Republican senators asked Commerce Secretary Elise Stefanik to water those rules down, citing lost jobs in Arizona and New York (Reuters). The result: harsh rhetoric up front, porous enforcement behind the scenes.
Geopolitically the approach is boomeranging. A Financial Times analysis of Trump’s new tariff package shows Southeast‑Asian exporters scrambling for alternatives. Vietnam’s tech ministry has since opened talks with Chinese foundry group CXMT on a $1 billion memory‑chip joint venture (local press, 18 April). Beijing, for its part, warned of “reciprocal measures” after the tariffs (Reuters)—while quietly offering venture finance to the very countries Washington hopes to woo.
Governance is where the stakes bite deepest. America’s civic DNA still offers safeguards neither Beijing nor Moscow emulate: freedom‑of‑information requests, investigative journalism, an adversarial court system. Those features make U.S. dominance potentially safer than an opaque alternative. But none are constitutional guarantees for AI; they are norms, and norms can buckle under partisan pressure. Trump has shifted the centre of gravity from “human‑centric” to “America‑first.” The next president could pivot again—and so on, whiplash without end.
If neither unilateral American control nor an unchecked Chinese surge looks stable, what does? Think of AI stewardship as shared custody:
- Lock in a global baseline. The OECD AI Principles and the UN’s nascent Algorithmic Safety Framework need the status—and teeth—of the Nuclear Non‑Proliferation Treaty. Ratification should be a pre‑condition for access to frontier models and advanced chips.
- Let regions add their own layers. The freshly enacted EU AI Act may read like Brussels bureaucracy but it offers continuity—something U.S. policy lacks. Singapore’s model algorithm guidelines do the same in miniature. Layered governance means one election can’t topple the whole edifice.
- Fund open‑science watchdogs outside any single jurisdiction. Red‑team eval suites, interpretability benchmarks and incident databases should be hosted by independent, multi‑sited consortia—financed via an OECD‑administered trust—to ensure no president or premier can pull the plug.
- Compete where competition helps. Talent, basic research, low‑risk commercial deployments. Shift from trying to choke rivals to raising the floor for everyone. Export controls should target specific military end‑uses, not every data‑centre GPU in Asia. Subsidised compute vouchers for allied universities would buy more goodwill—and intelligence—than tariff exemptions dangled like geopolitical candy.
Critics will call this naïve in a world of rising authoritarianism. Yet the alternative—technological primacy anchored to a single flag—breeds its own hubris. When the dollar’s dominance felt unassailable, Wall Street invented weapons of mass financial destruction. When the internet looked like an American fiefdom, surveillance capitalism filled the void. AI will be no different if left to the swagger of electoral swings.
Donald Trump is right about one thing: governing by precaution alone risks smothering innovation. But governing by bravado risks something worse—turning a general‑purpose technology into a nationalist totem. The world does not need America to win the AI race; it needs America, China and everyone else to run it on a track whose guard‑rails no leader can kick aside in a fit of election fever.
Until Washington binds itself as firmly as it seeks to bind Beijing, cheering for a U.S. victory means cheering for a moving target. And in a field where yesterday’s code can spawn tomorrow’s weapons, the target cannot be allowed to move.
