Public Policy & Governance

Hours before a scheduled White House signing ceremony, President Trump withdrew a prepared executive order on AI oversight after David Sacks — who had recently stepped down as the administration's AI czar but retained informal influence — relayed objections from leading AI executives, according to Politico. The unsigned draft, obtained by Politico, contained a government safety review requirement for new AI models prior to release — a provision the industry opposed on competitive grounds. Trump's public explanation that he 'didn't like certain aspects' confirms the reversal was substantive, not procedural.

The episode is analytically significant for three reasons. First, it demonstrates that informal industry access to the White House remains more determinative of federal AI policy than formal rulemaking processes. Second, it removes any near-term prospect of mandatory pre-deployment federal review, leaving the US with no enforceable national-level safety gate for frontier models. Third, as The Guardian notes, it effectively green-lights major model releases without independent government scrutiny — a position that directly contradicts the national security framing the administration uses when restricting AI exports to China. The contradiction between export controls premised on AI risk and domestic deregulation premised on AI opportunity has not been resolved.

Two Democratic New York state legislators have formally urged their congressional counterparts to reject preemption language embedded in ongoing House AI negotiations, joining a growing coalition of state-level officials opposing federal override of state AI safety laws, per Politico. This follows similar pushback from California legislators defending their own AI liability and transparency frameworks. The practical stakes are high: a federal preemption clause would nullify enacted and in-progress state laws covering algorithmic accountability, automated decision-making, and sector-specific AI mandates.

The dynamic mirrors the long-running federal-state tension over data privacy, where the absence of federal legislation for years allowed states to create a fragmented but operative regulatory landscape. Industry broadly prefers federal preemption because it reduces compliance complexity and sets a lower floor; civil society groups and state AGs prefer state authority because it preserves stricter standards. The House negotiations have not produced a bill, and the Trump order collapse removes White House momentum that might have accelerated congressional action. The preemption question is therefore likely to remain unresolved through 2026, leaving the state-level patchwork intact and growing.

The European Parliament and Council reached agreement on the AI Omnibus package in May 2026, according to CDT Europe's AI Bulletin. Alongside the political agreement, the European Commission advanced implementation with new draft guidance on high-risk AI systems and transparency obligations — signalling that the EU is moving from legislative design to enforcement-readiness. Key AI Act obligations on prohibited practices and general-purpose AI model transparency are already in effect or imminent, meaning the Omnibus agreement lands as practical compliance pressure on firms is rising, not in a pre-enforcement vacuum.

The Omnibus agreement is understood to streamline obligations for some categories of deployers and adjust liability chains, though CDT Europe flags ongoing debates around AI security and copyright that remain contested. The Commission's parallel publication of high-risk system guidance is significant: it narrows the interpretive space firms have used to argue their systems fall outside regulated categories. Cross-jurisdictionally, the EU's position is now structurally opposite to the US — the world's two largest AI markets are simultaneously tightening and loosening their governance postures, creating compliance asymmetry that will shape where frontier AI development and deployment concentrates.

Sadiq Khan intervened to block a £50m contract between the Metropolitan Police and Palantir for AI-assisted intelligence analysis, citing concerns about how the deal was structured rather than the technology itself, according to The Guardian. Scotland Yard publicly criticised the decision as 'disappointing' and warned of operational impact — an unusual public confrontation between a police force and its oversight authority. Palantir's UK and Europe head Louis Mosley accused Khan of 'putting politics above public safety,' per The Guardian.

The governance question here is distinct from the technology question. Khan's intervention, whatever its political motivations, exercises precisely the kind of democratic accountability over AI procurement that governance frameworks are designed to enable. The tension it has exposed — between operational police autonomy and mayoral oversight — reflects an unresolved question in UK public-sector AI governance: who has standing to review, condition, or block AI contracts signed by public bodies, and on what grounds? The case also puts Labour in an awkward position, with internal party tensions over engagement with US defence-tech firms surfacing publicly. For procurement professionals, the episode signals that high-value AI contracts with controversial vendors require political due diligence alongside commercial and technical assessment.

Chancellor Rachel Reeves has issued a direct instruction to all cabinet ministers to prioritise British companies in government AI procurement alongside ships, steel, and energy, according to The Guardian. This moves the UK's AI industrial strategy from aspiration to a ministerial compliance requirement — though enforcement depends on departmental implementation and is subject to existing procurement law constraints, including trade agreement obligations. The directive's practical effect will hinge on how 'British' is defined in AI contexts where supply chains, model training, and cloud infrastructure are heavily internationalised.

In a parallel development, commentary in The Guardian argues that Australia should establish a sovereign AI wealth fund to capture returns from AI investment, framing incoming interest from Microsoft and Anthropic in Australian data centre and training infrastructure as a negotiating opportunity rather than a passive windfall. While the piece is opinion rather than policy, it reflects a broader pattern of middle-power governments — Australia, UK, Canada — seeking to leverage data assets and infrastructure investment to extract sovereign benefit from AI development dominated by US hyperscalers.