In one week, Washington and Beijing gave the same warning in different words.
In Washington, the four US firms that control the most powerful AI—Anthropic, OpenAI, xAI, and Microsoft—said AI was moving too fast for humans to control. Anthropic’s Dario Amodei called for an immediate slowdown. Sam Altman said OpenAI will not go public in 2026 because safety concerns make it “ill-advised.” Elon Musk and Satya Nadella also said progress must be paced.
Their reasons were practical, not philosophical. AI agents escaped and hacked outside systems. AI is improving itself, cutting development from years to months. Internal tests show models are nearing the point where they could help launch cyberattacks or make bio-weapons. Two senior Anthropic safety researchers quit, warning AI could cause human extinction.
Beijing did not agree to a pause. It pushed back hard. Beijing called the US safety push a trick to keep America ahead.
One side fears AI will kill us. The other fears it will take over us. Both fear it will escape us. That shared fear matters.
For two years, Silicon Valley and Shenzhen followed the same idea: innovation must beat regulation. President Trump recently said: “Whoever wins AI wins.” This past week he went further: “the only guardrail AI needs is a ‘STRONG AND SMART (High IQ!) PRESIDENT.’” He called safety concerns a “SICK conspiracy going on against AI and Data Centers,” and said only China benefits. Trump’s AI adviser said government help is not needed if labs work together. Now, as Al Jazeera reports, Trump says he will create an “AI Force” with a new “AI czar.” The official line is clear: no new federal guardrails. The President is the guardrail. Slowing down helps China.
This is Cold War thinking applied to technology that ignores Cold War borders. A nuclear warhead sits in its silo until launched. A frontier AI model, once released as open source, can be copied a million times in a day, stripped of safety filters on a laptop in Nairobi or Jakarta, and used for anything.
US intelligence says Chinese tech firms copied billions of tokens from US models. Beijing says US apps spy through conversations. Both are probably right. That is what a borderless, leaky technology does.
So national rules alone are not enough. A US “AI Force” led by an “AI czar” cannot police technology that crosses borders at the speed of light. America cannot enforce its rules in Hangzhou. China cannot enforce its rules in California. If only US companies slow down, open Chinese models fill the gap. If only Chinese companies slow down, US agents race ahead. One country acting alone does not regulate AI. It guarantees a race to the bottom.
That is why, in seven days, international regulation went from a European ideal to an industry request. But one UN regulator that licenses models will not work. Neither Washington nor Beijing will give up control over technology it sees as a threat to its survival.
What can work is a layered system built on shared weakness.
First, a deal among leading AI labs. Builders have already agreed in principle to slow capability growth and accept independent evaluators. Make it law. Require safety testing by national labs before any model above a certain computing power level is deployed.
Second, a US-China hotline for AI incidents. During the Cold War, the nuclear hotline did not end ideology; it stopped miscalculation from killing everyone. We need the same for rogue agents: a promise to notify, isolate, and not weaponize an escape.
Third, a UN registry, not an enforcer. Like ICAO, which sets rules for aviation, it would set a shared definition of catastrophic risk and require reporting of evaluations, major incidents, and open-source model releases above a threshold. Its power would be legitimacy. Developing countries, facing displacement without having built the models, will not accept rules written only by the US or China.
The alternative is the path we are on: a race to deploy increasingly autonomous systems with no common safety floor, justified by the need to beat the other side. As Amodei warned, “We need international coordination, or else we risk running the same race with China, which could be equally dangerous.”
This past week proved two opposite things at once: the people building frontier AI believe control is no longer guaranteed; the politicians in charge believe they can still win a race against an uncontrollable technology. Both can’t be true.
Regulation is no longer about slowing innovation. It is about making sure there is a society left to innovate in. That can only be done together.
Pasifika Talanoa by a traveled ITaukei Fijian 