Public Policy & Governance

The Albanese government's new national AI plan, confirmed on July 19, goes beyond the Prime Minister's earlier speech by attaching enforceable constraints to how government departments and agencies use AI in automated decision-making. According to The Guardian, the plan is expected to extend to consumer protections, workplace safety, and privacy — and is accompanied by a push for digital duty of care legislation. This is a substantive legislative action, not a consultation or aspirational framework, which distinguishes it from most comparable G20 government AI initiatives to date.

The commentary from Julianne Schultz in The Guardian correctly identifies that the harder policy challenges — sovereign capability, security, and public benefit distribution — remain unresolved. The government's framing around data centres and copyright is at risk of being captured by industry interests before the deeper structural questions of accountability and democratic oversight are addressed. Implementation gaps will centre on enforcement mechanisms: which body audits government agency compliance, what remedies exist for affected citizens, and whether the duty of care legislation can pass a Senate where Labor lacks a majority.

The abrupt departure of Chris Fall, Director of the Commerce Department's Center for AI Standards and Innovation, removes the key architect of the White House's voluntary review process for frontier AI models at a politically sensitive moment. Politico reports Commerce is now seeking a replacement, but the vacancy underscores the structural fragility of an AI safety architecture built on voluntary commitments rather than statutory authority. Fall's role was central to operationalising the Biden-era executive order provisions that the Trump administration inherited and selectively retained.

The broader federal posture is defined by non-binding instruments: Trump's 'Ratepayer Protection Pledge', signed by approximately 200 entities, is explicitly non-binding per The Guardian, and the $5 billion AI research initiative reported by Politico is a funding steering exercise rather than a regulatory framework. Meanwhile, Senators Cruz and Blackburn met with Trump ahead of a Congressional markup on AI policy and children's online safety, per Politico, suggesting the legislative track is advancing separately from the executive branch's safety infrastructure — with no clear coordination mechanism between the two.

CDT Europe's formal response to the European Commission's consultation on draft guidelines for classifying high-risk AI systems identifies ambiguities that could allow providers and deployers to self-classify out of the Act's most demanding compliance tier. The guidelines are meant to clarify when an AI system qualifies as high-risk under Annex III of the AI Act — a determination that triggers the bulk of the regulation's obligations including conformity assessments, transparency requirements, and human oversight mandates. CDT Europe's published feedback argues the current draft provides insufficient clarity on boundary cases, which creates a de facto loophole for systems deployed in healthcare, employment, and education.

This consultation is a confirmed rule-making step, not a political statement — the Commission is required to finalise these guidelines to operationalise the Act, which began phased application in 2024. The implementation gap here is between the Act's legal text and the practical guidance regulators and compliance officers need to apply it. If the final guidelines do not address CDT Europe's concerns, enforcement by national market surveillance authorities will be inconsistent across member states, undermining the single market logic of the AI Act itself.

Andy Burnham's reported instruction to officials to draft plans abolishing the Department for Science, Innovation and Technology has provoked a cross-sector backlash from MPs, Whitehall officials, and industry, according to The Guardian. DSIT currently houses the AI Safety Institute, the government's technical AI safety capacity, and is the primary interlocutor with the EU on AI Act alignment questions. Abolishing it mid-cycle would disrupt the institutional knowledge base at precisely the moment the UK needs to decide whether and how to align with EU AI Act provisions or diverge toward a more permissive framework.

It is important to note that this is a reported proposal under development, not a confirmed decision — officials are drawing up plans, not implementing them. However, the political signal matters: a new Prime Minister publicly directing abolition studies creates uncertainty for ongoing regulatory processes, international partnerships, and the retention of technical staff within DSIT. The Chatham House analysis on lessons from the Kimi K3 and WAICO incidents separately argues the UK lacks the institutional mechanisms to respond rapidly to AI safety incidents — a gap that DSIT's abolition would deepen, not close.

A coalition of more than 20 companies including Microsoft, Meta, Nvidia, Dell, Palantir, and venture capital firm a16z have signed a joint letter defending open-weight AI models, shared exclusively with Politico ahead of the Congressional AI policy markup. The letter is a direct lobbying intervention designed to pre-empt regulatory restrictions on open-source model releases, which some legislators and safety advocates have argued pose proliferation risks. The coalition's composition — spanning hyperscalers, hardware manufacturers, enterprise software, defence-adjacent firms, and capital — signals an unusually broad industry alignment on this specific policy question.

The CDT's state-level analysis noted in their Tech Talks series provides relevant context: state legislatures are increasingly the venue where open-weight model restrictions are being debated, and industry has been less organised at that level. The federal lobbying effort may be partly designed to establish a pre-emptive federal standard that supersedes state-level restrictions — a strategy that mirrors the industry's successful pre-emption campaigns in other technology regulatory domains.