Public Policy & Governance

Senator Kennedy's planned 'kill switch' bill — to be introduced Wednesday under unanimous consent — is notable as a concrete legislative action rather than a committee discussion or white paper. However, the unanimous consent procedure means any single senator can block it, and the absence of detail on the bill's substantive provisions makes its enforceability impossible to assess at this stage. It sits within a Congress where AI legislation has repeatedly stalled, and where the procedural vehicle signals speed over coalition-building. Politico

Complicating the legislative picture further, Representative Frank Pallone, the top Democrat on the House Energy and Commerce Committee, has pushed back against establishing an AI select committee. This is significant: Pallone would be a central figure in AI legislation if Democrats retake the House in November, and his opposition to a dedicated committee suggests he prefers AI governance to run through existing committee jurisdiction rather than a new institutional structure. His positioning now shapes the post-midterm legislative architecture. Meanwhile, former President Obama's reported closed-door call for a Democratic AI safety framework adds political weight but is not a concrete policy action — it is elite signalling, not lawmaking. Politico

OpenAI's direct appeal to UK ministers to legislate AI safety — framed as seizing a 'political window' — is a strategically calibrated move. Coming immediately after Anthropic's own calls for AI slowdown and its separate appeal to the same UK parliamentary committee, it represents two leading AI labs urging government intervention on their own industry. Policy professionals should treat this with analytical scepticism: companies advocating for regulation frequently do so when they calculate that a specific regulatory framework would entrench their market position or impose higher compliance costs on smaller competitors. The Guardian

Microsoft's publication of a voluntary 'code of conduct' for AI model training, by contrast, is a self-regulatory instrument with no enforcement mechanism — its publication by CEO Mustafa Suleyman on social media rather than through a regulatory filing underlines its political rather than compliance character. The UK government has not yet announced a legislative timetable in response to these industry appeals, meaning the gap between the lobbied-for regulation and any enforceable law remains wide. The cross-party parliamentary committee's warnings on human rights threats suggest legislative interest, but committee reports do not constitute law.

ECB President Lagarde's warning that Europe must develop domestically run AI infrastructure to avoid 'unprecedented leverage' from the US or China in trade negotiations represents a material escalation in the EU's strategic framing of AI governance. This is not aspirational innovation policy — it is a central bank president framing AI dependency as a financial stability and geopolitical risk. The ECB's institutional weight gives this a different register than similar statements from the European Commission. The Guardian

The statement arrives as the EU AI Act's implementation timelines are active, but enforcement capacity across member states remains uneven. Lagarde's framing — that European AI models need only be 'good enough' for most tasks and run from domestic datacentres — implicitly endorses a sovereignty-first approach over a capability-maximisation one, which has direct implications for procurement rules, state aid frameworks, and the regulatory treatment of non-EU AI providers operating in European markets.

The New Mexico Supreme Court's decision to fine and hold in contempt attorney Stephen Aarons for submitting an AI-generated brief containing fabricated testimony and invented witnesses is the most concrete enforcement action in the current news cycle directly relevant to institutional AI governance. This is a court, not a legislature, acting — and it establishes that existing professional conduct obligations extend to AI-generated content without requiring new AI-specific legislation. The Guardian

The case has direct implications for public sector legal practice, where AI-assisted document drafting is increasingly common. Courts across the US have issued standing orders requiring disclosure of AI use following prior hallucination cases, but enforcement through contempt — as opposed to reprimand — raises the compliance stakes materially. For government legal departments and regulatory agencies that produce court filings, this ruling reinforces that verification obligations are not aspirational but judicially enforceable.

The public warning from Australian Signals Directorate chief Abigail Bradshaw that legacy government systems face acute vulnerability to AI-enabled cyberattacks is a significant intelligence agency disclosure in the context of active national AI governance negotiations. It adds operational urgency to a debate that has largely proceeded at a policy-design level. Australia currently lacks binding AI legislation — a gap highlighted separately by commentary arguing the absence of a federal human rights act compounds the risk of algorithmic decision-making in public services. The Guardian

The dual signals — offensive AI threat to government infrastructure and the absence of defensive legal frameworks governing AI deployment in public sector decision-making — represent distinct but compounding governance failures. Bradshaw's comments are a policy input to the government's ongoing AI guardrails negotiation, but the 'enormous' cost of system modernisation she cited suggests that even if legislative frameworks are agreed, implementation infrastructure lags significantly behind.