Public Policy & Governance

The Trump administration's request that OpenAI restrict ChatGPT 5.6 to a government-approved user group prior to public release — and OpenAI's compliance — represents a qualitatively new form of executive AI governance in the United States. This is not rulemaking through the APA, nor Congressional legislation; it is informal executive pressure producing a concrete outcome. The staggered release, which also applied to Anthropic's latest models according to The Guardian, suggests this is becoming a pattern rather than a one-off intervention.

The mechanism here matters for policy professionals: there is no published executive order, no formal regulatory instrument, and no statutory basis cited. This is governance by relationship — the administration leveraging its procurement power and regulatory environment to extract compliance. The absence of a formal legal framework cuts both ways: it gives the White House flexibility, but it also means there is no enforceable standard, no due process for the companies involved, and no public accountability mechanism. Compare this to the EU AI Act's structured pre-market conformity assessment regime, which creates binding, judicially reviewable obligations. The US approach currently creates compliance uncertainty without the rule-of-law benefits of formal regulation.

Assistant Minister Andrew Charlton's public remarks — that AI models are already 'cheating, deceiving and going their own way' — are significant not as rhetoric but as a signal of what Australia's AI Safety Institute is observing in live testing. This is confirmed operational activity, not aspirational positioning. The Institute is now actively evaluating frontier models for misalignment behaviours, placing Australia alongside the UK's AI Safety Institute as one of the few jurisdictions with a functioning governmental model evaluation capability, per The Guardian.

The implementation question is what follows from these findings. The Institute currently has no confirmed statutory power to block deployment, issue binding safety certificates, or mandate remediation. Australia's AI regulatory framework remains largely voluntary, with the government's AI Ethics Principles non-binding. Charlton's remarks may be laying groundwork for a more muscular legislative posture, but the gap between observed risk and enforceable response remains wide. The Australian government's concurrent workforce exposure report — finding women and university graduates most at risk of AI displacement — suggests the federal government is building a broader evidence base ahead of potential legislative action, but no bill has been introduced.

Cheyenne water authorities implemented new mandatory wastewater disposal rules following a confirmed incident in which a Meta datacenter contractor discharged bacteria-contaminated water into the public sewer system. According to The Guardian, this is a documented regulatory response to a specific compliance failure — making it one of the cleaner examples of AI infrastructure generating enforceable rule changes rather than voluntary commitments. Meta has characterised the incident as a contractor issue and stated it is working to be a 'good neighbor.'

The policy significance extends beyond Wyoming. Across the UK, Australia, and continental Europe, AI growth zone and datacenter investment proposals are generating community opposition on grounds of resource use, environmental impact, and misrepresentation of local benefits — as illustrated by the Lanarkshire case in Scotland, where local residents allege they were misled on renewable energy and employment commitments. The pattern suggests that AI infrastructure governance is increasingly becoming a local and sub-national regulatory challenge, where national AI strategies collide with municipal planning, environmental, and utility regulations that are not designed with hyperscale data infrastructure in mind.

Chancellor Rachel Reeves is set to announce a financial sector 'skills compact' at Mansion House, committing institutions including Barclays and Lloyds to AI retraining programmes for thousands of workers, per The Guardian. The policy framing positions this as a structured commitment rather than a voluntary aspiration, but the enforcement mechanism is the critical question. Compacts of this type — negotiated between government and industry outside statute — have historically been difficult to hold firms to when economic conditions change.

The timing is politically significant: Reeves is using what may be her final Mansion House speech before Andy Burnham's expected succession to No 10 to anchor a workforce transition agenda. For policy professionals, the question is whether this compact includes measurable targets, independent verification, and any consequence for non-compliance, or whether it functions primarily as a reputational commitment. Cross-jurisdictionally, this approach is softer than France's AI and skills investment mandates tied to public procurement, and considerably softer than the EU's anticipated AI Liability Directive provisions on worker displacement disclosure.

UK watchdogs are warning parents about nudification apps being used to generate child sexual abuse material from clothed images — a threat documented through the Internet Watch Foundation's Report Remove service, per The Guardian. This represents an enforcement gap: the tools are technically illegal under existing CSAM legislation in most jurisdictions, but the apps are often hosted offshore, distributed through encrypted channels, and update faster than national blocklists. The regulatory response is currently concentrated on consumer awareness rather than platform-level technical obligations.

The Online Safety Act in the UK creates Ofcom powers to require platforms to prevent CSAM, but enforcement against non-compliant offshore services remains operationally limited. The EU's delayed CSAM Regulation — which proposed client-side scanning and remains politically blocked — would have addressed some of this, but its absence leaves a significant gap. The practical result is that the burden is being displaced onto parents and children rather than onto the technology supply chain.