Version 2

The Trump administration postponed a planned AI security executive order on May 21, 2026, with Trump stating he didn't want regulation to get 'in the way of that leading' [1][2] — framing deregulation as a US-China competitiveness imperative. California Governor Newsom signed a 'first-of-its-kind' executive order the same day, treating AI-driven job displacement as a public policy problem and directing state agencies to study workforce protection mechanisms [6][7]. A December 2025 White House executive order explicitly aimed at eliminating state AI law obstruction [12], combined with a reported forthcoming 'ONE RULE' federal preemption order [13], reveals the federal position is not simple deregulation but an effort to block state governance while declining to govern federally. Legal analysts and advocacy groups are now challenging whether federal preemption without any federal governance alternative is constitutionally or democratically defensible [15][18].

The combination of federal safety deregulation and active federal preemption of state laws could produce a genuine governance vacuum — no level of US government constraining AI's workforce or security risks. If the Trump administration successfully prevents states from acting while continuing to postpone federal oversight, the period ahead is one where AI's economic and safety impacts accumulate without any public accountability framework at any level of government.

American AI governance in mid-2026 is shaped by a three-way standoff between federal deregulation, state-level intervention, and federal efforts to preempt state action. On May 21, 2026, the Trump administration postponed the signing of a planned AI executive order just hours before it was scheduled. Trump's own words — reported across the Associated Press, TechCrunch, the Wall Street Journal, and the New York Times — framed the delay as a competitiveness decision: 'I don't want to get in the way of that leading' [1][2][3]. The White House had been signaling this direction for weeks, with Politico reporting in early May that the administration was explicitly distancing itself from tighter AI oversight [4]. A prediction market opened the same day asking whether Trump would sign any delayed AI security order before August 21, 2026 [5].

California Governor Gavin Newsom moved in the opposite direction on the same day, signing what his office described as a 'first-of-its-kind' executive order treating AI-driven job displacement as a public policy problem rather than a corporate responsibility [6][7][8]. The order directs state agencies to study severance pay, subsidized employment programs, and other workforce support mechanisms for workers displaced by AI [6][9]. This builds on a multi-year pattern of California AI governance, including a September 2023 executive order preparing the state for AI broadly [10] and an April 2026 order on AI procurement standards for state agencies [11].

What complicates the simple federal-retreat-versus-state-advance framing is a December 2025 White House executive order — formally titled to 'eliminate state law obstruction of national artificial intelligence policy' [12]. StateScoop reported that a state AI law moratorium was omitted from the 2026 defense bill but that Trump is preparing a separate 'ONE RULE' executive order to achieve a similar preemptive effect [13]. New state AI laws that had taken effect January 1, 2026 [14] now exist in legal limbo as legal analysts examine the landscape and limitations of this federal preemption push [15][16][17]. Critics argue that preemption without a federal governance alternative is not governance at all — it simply removes accountability at every level [18]. The administration's configuration is thus unusual: simultaneously refusing to regulate AI federally while working to prevent states from filling that gap.

Running beneath these political battles is a substantive debate about which frameworks are even appropriate for AI governance. The AI Snake Oil critique of nonproliferation-style controls — that AI lacks the physical bottleneck of enriched uranium and that frontier capabilities can be replicated by nation-states within months of any given breakthrough [19] — has been amplified by additional policy and scholarly sources questioning whether the nuclear model fits AI at all [20][21]. The resilience-based alternative, emphasizing biosecurity screening, AI-assisted red-teaming, and infrastructure hardening [22][23][24], requires polycentric cross-agency coordination that is harder to achieve through executive action than either regulation or deregulation. On the US-China competition that Trump invoked to justify his delay, a USCC analysis described how China's open AI strategy reinforces its industrial dominance through a 'two loops' approach [25], while Stanford's 2026 AI Index continued tracking the capability gap [26] — data points available to both sides of the regulatory debate.