Public Policy & Governance

On 14 April 2026, the Trump administration's Office of Management and Budget published an inventory disclosing 3,611 active or planned AI use cases across the federal government — a figure 70% higher than the equivalent disclosure made in the final year of the Biden administration. The inventory, surfaced via a Guardian commentary by Nathan Sanders and Bruce Schneier, covers a range of sensitive governmental functions being delegated to AI systems, from benefits adjudication to law enforcement support. The disclosure itself was described as 'quiet,' with no accompanying press release or congressional briefing, reinforcing concerns that scale of deployment is outpacing any oversight architecture.

The gap between what has been announced and what is enforceable is significant. The Biden-era Executive Order 13960 established a framework for federal AI transparency that mandated these inventories, but it imposed no substantive constraints on deployment or requirements for independent audit. The Trump administration has not replaced that framework with anything more rigorous — it has simply inherited and accelerated the deployment pipeline while rhetorically deprioritising regulatory friction. There is currently no statutory requirement compelling agencies to publish impact assessments, obtain external review, or notify affected populations before deploying AI in consequential decisions. Congress has not moved binding legislation on federal AI procurement standards.

The Trump administration's positioning as a deregulatory force on AI has hit an early and visible contradiction. Within roughly two weeks of articulating a light-touch oversight approach, the administration imposed new export restrictions on Anthropic's newest AI model, prompting emergency meetings between senior White House officials and the company to negotiate what Politico describes as a 'truce.' One White House official signalled resolution would take more than a few days. The restrictions appear to stem from national security and export control considerations rather than safety regulation — but the practical effect is that a company operating under the expectation of a permissive federal environment is now facing government-imposed deployment constraints.

The episode is analytically significant because it illustrates the structural tension within the administration's own AI policy framework: the national security apparatus — Commerce Department export controls, likely BIS involvement — operates on a separate track from the economic deregulation agenda being driven by other White House factions. Industry advocates quoted by Politico are explicitly worried this signals the laissez-faire posture is not durable policy but a rhetorical position vulnerable to override by security agencies. This is not a dispute about AI safety regulation; it is a dispute about who controls AI policy inside the executive branch.

Access Now, Amnesty International, and more than 200 co-signatories have issued a joint statement calling for an immediate halt to the use of AI systems in military kill chains, framing AI-accelerated warfare as a mechanism for 'rubber-stamping killing at unprecedented speed and scale.' The statement explicitly invokes international criminal law, human rights law, and international humanitarian law as frameworks being undermined by current deployment practices. This is not an academic position: it references ongoing conflicts where AI-assisted targeting systems have been operationally deployed by state actors, including in contexts where post-hoc accountability for civilian harm has been absent.

The governance gap the coalition is targeting is real and largely unaddressed at the multilateral level. The UN Secretary-General's call for a legally binding instrument on autonomous weapons has not produced a treaty process with meaningful state participation. The U.S., UK, Israel, and other major military AI developers have resisted binding constraints, preferring non-binding principles and national implementation frameworks. The EU AI Act explicitly excludes military applications from its scope. The practical result is that the fastest-moving application domain for AI — military — is the least governed, and the coalition's statement is designed to generate political pressure ahead of any potential UN General Assembly or Human Rights Council action.

California's legislature is actively working through a genuine policy dilemma on AI-driven job displacement, with Politico reporting that Sacramento's current and incoming leadership face a choice between hard protective measures — potentially including restrictions on AI deployment in specific sectors — and softer managed-transition approaches such as retraining mandates, displaced worker funds, or enhanced unemployment insurance. California has the largest state economy in the U.S. and a history of setting regulatory precedents that propagate nationally, making this a higher-stakes deliberation than comparable efforts in smaller states.

What distinguishes this from political rhetoric is that the dilemma is concrete and fiscal: protective measures risk opposition from the state's large technology sector and may face preemption arguments if federal AI legislation eventually passes, while transition-support frameworks require sustained budget commitments in a state already managing significant fiscal pressures. No legislation has yet passed on this specific question, but the framing of the debate — protection versus mitigation — will shape what goes to a vote.