Public Policy & Governance

President Trump on September 19 announced the creation of an 'AI Force' and the appointment of an AI czar, providing almost no structural, jurisdictional, or budgetary detail for either initiative. The announcement came as Trump simultaneously rejected industry and bipartisan calls to slow AI development, repeatedly invoking China as the justification — 'We're leading China in AI' — framing the governance question as a competitiveness issue rather than a safety one. As Politico reported, tech sector representatives view the announcement as a response to mounting public concern rather than a considered policy design.

Critically, the House Speaker Mike Johnson cancelled scheduled votes and sent lawmakers home early ahead of midterm recess, described by The Guardian as occurring 'amid AI regulation frenzy.' This means Congress is exiting the session without passing AI-specific legislation, leaving the federal regulatory field to executive action of uncertain scope. A Politico poll found 63 percent of Americans believe AI poses at least a moderate risk of destroying humanity and a plurality favour pausing advanced model development — a significant political pressure point that the 'AI Force' announcement appears designed to absorb without committing to enforceable constraint.

A lawsuit filed against Anthropic, OpenAI, SpaceX AI, and Google alleges that their coordinated public statements calling to 'pace the frontier' of AI development constitute illegal coordination between competitors in violation of antitrust law, as reported by Politico. The legal theory is significant: it attempts to use competition law to prevent the dominant players from collectively managing the pace of a market in which they hold leading positions. If the claim has merit, it would effectively prohibit the industry's primary voluntary self-regulatory mechanism — joint commitments on development pace — as anticompetitive.

This creates a structural paradox for policymakers. Governments seeking voluntary industry cooperation on safety norms may find such cooperation legally untenable under existing antitrust frameworks, strengthening the case for legislated safe harbours or formal regulatory co-ordination bodies with antitrust immunity. The lawsuit also tests whether Dario Amodei's slowdown proposal, endorsed by Sam Altman and Elon Musk per The Guardian, was a genuine safety measure or market coordination dressed in safety language — exactly the suspicion critics raised publicly.

Governor Gavin Newsom issued an executive order on AI, which the EFF welcomed as opening a 'needed, thoughtful conversation' while immediately flagging that the order focuses on longer-term and speculative risks rather than the 'most immediate and current concerns' facing Californians. EFF's qualified support is a meaningful signal: the organisation is broadly skeptical of AI regulation that serves incumbents, so endorsement-with-caveats indicates the order is directionally defensible but operationally incomplete.

California's action follows its established pattern — SB 1047's veto by Newsom in 2024 and subsequent narrower executive action — of advancing the policy conversation without enacting binding compliance obligations. In the current federal vacuum, California's executive orders function as agenda-setting instruments for other state legislatures and, historically, as templates for eventual federal rulemaking. However, an executive order carries no legislative durability and can be reversed by the next governor, limiting its value as a compliance anchor for businesses or public agencies.

Tasmania's Justice Department confirmed a formal review of AI use in parole decisions after a parole condition imposed on convicted killer Susan Neill-Fraser was ruled invalid — the board had cited a non-existent case, almost certainly an AI-generated hallucination, in its legal reasoning. As The Guardian reported, the review was announced four days after the error became public, suggesting reactive rather than proactive governance.

This case is the most concrete example in the current news cycle of AI governance failure with direct legal consequences — it affected an individual's liberty conditions and invalidated a quasi-judicial decision. It is structurally distinct from speculative existential risk debates: it has occurred, it has a named victim of procedural failure, and it exposes the absence of procurement and usage standards for AI tools in justice systems. Common law jurisdictions including the UK, Australia, New Zealand, and Canada lack uniform standards for AI-assisted legal decision-making, and this case provides the evidentiary basis that reform advocates will cite in pushing for mandatory human review requirements.

Both The Guardian's analysis and Chatham House's expert commentary converge on a critical governance problem: US AI safety arguments are inseparable from competitive strategy toward China, and Beijing is explicitly rejecting the slowdown framing as a US attempt to lock in existing advantage. Chatham House warns that Anthropic's warnings should not serve as 'cover for entrenching US labs or escalating rivalry with Beijing,' while The Guardian documents that Amodei and others frame Chinese AI superiority as equally threatening as superintelligence risk.

This dynamic is not merely rhetorical — it directly constrains the multilateral governance options available. Institutions like the UN, ITU, or a proposed international AI safety body cannot function if the two leading AI powers frame the governance debate as a zero-sum competition. The Bletchley Process and Seoul AI Safety Summit produced non-binding declarations; without US-China alignment on even definitional frameworks, binding international instruments remain structurally blocked. Trump's explicit China-competition justification for rejecting the slowdown as reported by The Guardian forecloses the bilateral dialogue that would be prerequisite to any multilateral agreement.