Version 1

US AI governance is fracturing along a federal-versus-state axis: the Trump administration postponed a planned AI executive order, citing fears that regulation could slow American AI companies competing with China [1], while California Governor Newsom moved in the opposite direction by signing what is described as the first executive order explicitly treating AI-driven job displacement as a public policy problem [2]. Running parallel to both moves, academic researchers at AI Snake Oil published a substantive critique arguing that neither heavy restriction nor nonproliferation-style controls are the right response to AI risks, and that societal resilience is a more durable alternative [3].

The divergence between federal deregulation and state-level intervention sets up a fragmented US governance landscape where major policy decisions on AI's economic and security impacts may be made piecemeal by states rather than through coherent national frameworks. If the AI Snake Oil critique holds — that nonproliferation controls are unenforceable for AI in a way they are not for nuclear weapons — then both the federal retreat and state-level restrictions may be addressing the wrong problem entirely.

American AI policy in mid-2026 is being pulled in two directions simultaneously. At the federal level, the Trump administration stepped back from what had been a planned AI executive order, with Trump explicitly framing the delay as a competitiveness decision: the US is currently leading global AI rivals, and he did not want regulation to jeopardize that position [1]. The postponement reflects a White House calculus that prioritizes speed and market freedom over precautionary frameworks, and fits a broader pattern of treating AI primarily as an economic and geopolitical asset rather than a technology requiring governance guardrails.

At the same time, California Governor Gavin Newsom signed an executive order that takes the opposite premise as its starting point: that AI-driven job displacement is a public policy problem, not merely a corporate responsibility [2]. The order directs state agencies to study severance pay, subsidized employment programs, and other workforce support mechanisms. Described as the first of its kind, it positions a major US state as a de facto regulator in a domain where the federal government is deliberately stepping back — a dynamic that could intensify if other states follow suit and begin diverging in their approaches.

The intellectual backdrop to both moves is contested. A detailed essay by Sayash Kapoor and colleagues at AI Snake Oil argued on the same day against what they call 'extraordinary' government interventions — measures that impose costs on companies not directly responsible for AI misuse and that bypass democratic governance processes [3]. Their core claim is that AI nonproliferation, unlike nuclear nonproliferation, lacks an enforceable physical bottleneck: the core techniques for building frontier AI are widely known, and nation-states can match frontier capabilities within months of any given breakthrough [3]. Export controls and voluntary pre-deployment evaluations buy only months of delay before capabilities become broadly accessible via open-weight models [3]. The alternative they advocate — resilience through AI-assisted red-teaming, biosecurity screening, and infrastructure hardening — addresses misuse risks without restricting beneficial access, but requires coordination across many agencies rather than the kind of unilateral executive action that is politically easier to execute [3].

Taken together, these three developments sketch a governance landscape where the federal government is retreating for competitive reasons, one large state is advancing for labor-market reasons, and a credible academic voice is questioning whether either approach addresses the actual risk landscape. The result is a policy vacuum at the national level that states and critics are each, in different ways, rushing to fill.