On Sunday, President Trump publicly downplayed calls to slow the development of artificial intelligence, framing safety warnings as the product of “negative forces” who misunderstand America’s competitive position relative to China. The remarks came within hours of multiple major outlets reporting that Anthropic’s CEO Dario Amodei had called for AI development to slow down, and that a former Anthropic researcher told the BBC he and colleagues were “genuinely frightened” for humanity’s future. Separately, the UK government rejected a proposed “kill switch” for dangerous AI systems, citing concerns about competitiveness. The convergence of these events in a single weekend illustrates, with unusual clarity, the governance problem at the heart of the AI moment.
The received wisdom
The progressive-technocratic framing treats this as straightforwardly a case of political negligence in the face of expert consensus. The scientists and engineers closest to the technology are raising alarms with unusual candour; politicians — especially on the American right — are dismissing those alarms for reasons of competitive positioning and ideological hostility to regulation. The parallel drawn is to early climate change, when the fossil-fuel industry’s political allies successfully delayed action by decades. On this reading, the correct response is clear: international regulatory frameworks, mandatory safety testing, and public accountability for AI labs analogous to the FDA approval process for pharmaceuticals.
This framing has real force. The Anthropic researcher’s estimate of greater than ten percent probability of AI causing mass human casualties is not a fringe position — it represents a significant cohort of credentialled AI researchers who are not conspiracy theorists. The governance gap is real.
A different read
But the technocratic-regulatory response carries its own complications that the mainstream framing tends to elide. The first is that “slow down” is not a policy — it is a gesture. Amodei’s call for deceleration was accompanied by no concrete proposal for who would enforce it, at what technical threshold, and with what verification mechanism. The history of voluntary industry commitments to self-regulation in technology is not encouraging: social media companies made extensive voluntary commitments to content moderation, and the results were… mixed. A genuine regulatory framework requires an enforcement agency with technical capacity, and no such agency currently exists anywhere in the world at the frontier of AI development.
The second complication is the geopolitical one, and here Trump’s framing — though crude — is not entirely wrong. If the United States were to impose meaningful AI development constraints unilaterally, the primary beneficiary would be the Chinese state, which operates under no such constraints and has made AI development a declared national priority. The arms-control analogy is instructive: the Non-Proliferation Treaty worked, imperfectly, because the major nuclear states had a shared interest in preventing proliferation to third parties. An AI deceleration regime would require the United States and China to agree that neither would race ahead — an agreement with essentially no precedent in the history of strategic competition between great powers.
The third complication is more uncomfortable still: the safety warnings are coming from inside the houses that benefit most from fear-based regulation. When Anthropic’s CEO calls for a slowdown, one should at least pause to ask whether the effect — if not the intent — is to entrench current frontier labs against competitive entry. Regulatory capture is not a conservative fantasy; it is a documented phenomenon in every major regulated industry. The pharmaceutical sector, which is frequently cited as the model for AI regulation, is also the sector whose regulatory framework has been most systematically colonised by incumbent manufacturers.
None of this means the warnings should be dismissed. A 10-plus percent probability of existential harm, if credible, would justify almost any precautionary measure. But the choice is not between “regulate” and “ignore.” The more intellectually honest framing would be: what specific capabilities, at what specific technical threshold, create what specific pathways to harm, and what interventions would actually reduce those pathways without simply shifting the risk to less transparent jurisdictions? That is a hard question, and nobody in the current debate — not Trump, not Amodei, not Whitehall — is squarely answering it.
What the weekend’s events actually reveal is a bipartisan governance vacuum. The American right won’t regulate because it fears Chinese competition. The British government won’t regulate because it fears being left behind economically. The labs won’t self-regulate because the incentives run entirely in the other direction. And the left’s proposal — “slow down,” with no enforcement mechanism — is a slogan rather than a strategy. In the meantime, the technology proceeds at its own pace, indifferent to the political calendar.
What to watch
Watch whether the UK government’s rejection of the “kill switch” proposal generates any serious parliamentary pushback, or whether it passes without consequence — the latter would confirm how thin the political appetite for AI governance actually is. Watch also whether Trump’s dismissal of the safety lobby is followed by any concrete deregulatory action, such as reversing Biden-era executive orders on AI testing requirements. And watch Anthropic: a company whose CEO is publicly calling for slowdown while simultaneously raising multi-billion-dollar funding rounds is navigating a tension that will eventually become untenable.
— J