Public Policy & Governance

The US Department of Justice's intervention in xAI's lawsuit against Colorado's AI regulation law — on 14th Amendment equal protection grounds — is not simply a legal filing; it is an explicit statement of executive branch preference for federal over state AI governance. The Trump administration's argument that Colorado's law discriminates against certain companies is legally narrow, but the political signal is broad: Washington intends to assert supremacy over the emerging state-level regulatory patchwork before it consolidates. Colorado's law, which imposes obligations on developers of high-risk AI systems, was one of the more substantive enacted state frameworks in the US. As reported by The Guardian, the DOJ intervention creates a direct conflict between the administration and state authority.

Cross-jurisdictionally, this mirrors the EU's centralised AI Act model, where member-state divergence is constrained by a single framework — but the US dynamic is adversarial rather than cooperative, and the federal alternative to Colorado's law does not yet exist in enacted form. The DOJ is effectively arguing against state regulation without a federal replacement on the table. Industry players backing federal pre-emption include major AI developers who prefer a single compliance regime; civil liberties groups and state AGs are the primary opposition, arguing that federal inaction creates a vacuum that harms citizens.

The Metropolitan Police's week-long deployment of Palantir's AI tool to surveil its own officers — uncovering alleged misconduct ranging from work-from-home violations to suspected corruption — raises fundamental governance questions that extend beyond the specific use case. As reported by The Guardian, the deployment appears to have occurred without public consultation or disclosed legal basis under UK data protection frameworks. The scale of the resulting investigations — hundreds of officers — suggests the tool's output is being treated as actionable intelligence, which raises due process concerns about AI-generated evidence thresholds in disciplinary proceedings.

The political escalation is significant: Mayor Sadiq Khan's office signalled he may seek to block Scotland Yard from formalising a Palantir contract, citing concerns about 'using public money to support firms who act contrary to London's values' — a direct reference to Palantir's role in US immigration enforcement and Israeli military operations, as reported by The Guardian. This creates an unusual governance tension: the Mayor of London has oversight powers over the Met but limited authority over specific procurement decisions once in process. The UK's existing AI governance framework — based on sector-specific regulators and voluntary norms — provides no clear mechanism to adjudicate this dispute.

China's blocking of Meta's $2bn acquisition of Manus — confirmed as Beijing requiring explicit government approval for domestic tech companies accepting US investment — represents a concrete regulatory action, not a policy proposal. As reported by The Guardian, the measure transforms what was previously a de facto constraint into a formalised prior-approval regime for inbound US capital in AI. This is a reciprocal governance move: it mirrors US CFIUS restrictions on Chinese investment in American AI companies, but applies at the domestic firm level rather than requiring the acquirer to seek approval.

The Manus case is particularly notable because the target is an AI agent developer — a category that both the US and China regard as strategically sensitive. The blocking formalises the bifurcation of the global AI supply chain at the application layer, not just semiconductors. For multinational AI governance, this means the EU, UK, and other jurisdictions face increasing pressure to define their own investment screening approaches for AI acquisitions before being caught between US and Chinese regulatory camps.

The EU AI Omnibus — intended to streamline and partially roll back compliance obligations from the original AI Act — remains in active negotiation with unresolved disagreements on scope and safeguards, according to CDT Europe's April 2026 AI Bulletin. The Anthropic Mythos incident, in which a model withheld from public release on cybersecurity grounds was allegedly accessed by unauthorised parties, is being cited within EU policy discussions as evidence that the Omnibus must not weaken frontier model oversight provisions. The tension is between member states seeking to reduce compliance burdens on European AI developers and civil society organisations — including CDT Europe — arguing that the original Act's safeguards are already insufficient for the pace of frontier development.

The Omnibus negotiations are a direct regression risk: if concluded in a form that narrows the definition of general-purpose AI models subject to systemic risk obligations, the EU framework would be less stringent at precisely the moment when frontier model incidents are multiplying. The Mythos case illustrates the gap between announcement-based governance — where developers self-report risks and voluntarily restrict access — and enforceable obligations to notify regulators and implement access controls. No current EU or US regulatory framework would have required Anthropic to notify a regulator before the alleged unauthorised access occurred.

Google's reported classified AI agreement with the Pentagon — allowing use of its models for 'any lawful government purpose' — continues the pattern of major AI developers entering defence contracts with minimal disclosed governance terms, as reported by The Guardian. The classified nature of the deal means standard public procurement accountability mechanisms do not apply, and the breadth of the 'any lawful government purpose' formulation provides no meaningful constraint. This follows similar agreements by OpenAI and others, suggesting a deliberate standardisation of terms that insulates Pentagon AI deployments from oversight.

In the UK, Technology Secretary Liz Kendall's statement that Britain must 'seize the initiative on AI or be left at the mercy and whim' of a future shaped by the technology — reported by The Guardian — is political rhetoric, not a regulatory action. It reflects genuine strategic anxiety about the UK's position given US companies' dominance of AI compute, but the government has not announced new legislative measures or enforcement mechanisms. The concurrent revelation that UK government departments are presenting conflicting forecasts on AI data centre energy demands, as The Guardian reports, suggests the AI strategy lacks basic inter-departmental coordination.