Public Policy & Governance

A bipartisan House proposal would require leading AI companies to undergo mandatory third-party safety assessments, and OpenAI has formally endorsed the measure — a notable strategic shift for a company that has historically resisted hard regulatory mandates. OpenAI has also endorsed three separate bills targeting AI-amplified biosecurity threats, including standardization of biological data. The dual endorsements signal that at least one major lab has calculated that participating in regulatory design is preferable to having it imposed externally. Politico

However, critics — including civil society organizations and some legislators — are raising credible concerns about regulatory capture: the risk that audit frameworks designed in collaboration with the very companies being audited will be structured to minimize compliance burden rather than surface genuine risk. Congress has no enacted AI legislation to date. Senators Sanders and Blumenthal are pushing for urgency, with Sanders having announced legislation to ban 'superintelligent' AI outright, but neither chamber has moved a bill to the floor. The proposals remain at consultation and endorsement stage, not law. Politico

Leaked government documents reveal the Albanese government is actively considering an opt-out model for AI training data, which would make all publicly available Australian content — including creative works, photographs, and small business websites — accessible to AI companies by default unless rights holders affirmatively opt out. This represents a fundamental inversion of Australian copyright law, which currently requires consent before use. The proposal emerged as senior OpenAI personnel met directly with Labor ministers, and reports indicate OpenAI warned that Australia risks missing out on data centre investment without regulatory accommodation. The Guardian

The political backlash is immediate. Crossbench Senator David Pocock has accused Labor of throwing creatives 'under the bus', and creative industry groups are warning the opt-out mechanism is practically unworkable — the burden of monitoring and objecting to use at scale falls on individual rights holders rather than AI companies. This mirrors similar debates in the EU, where the AI Act's text mining exception includes opt-out provisions but has faced implementation criticism. The Australian proposal is not yet law — it remains at the leaked-consultation stage — but the direction of travel is clear, and the political window for course correction is narrowing as investment negotiations with U.S. tech firms advance. The Guardian

The divergence between OpenAI and Nvidia on AI regulation is now explicit and public. At a San Francisco conference, Nvidia CEO Jensen Huang reiterated his position that 'we don't need any new laws', directly echoing the Trump administration's deregulatory stance. Anthropic's Dario Amodei called for a development slowdown using automotive safety analogies. OpenAI's Sam Altman called for more rigorous security measures. This is not a marginal disagreement — Nvidia's business model is hardware sales volume, structurally incentivizing acceleration, while foundation model companies face existential liability from safety failures and have a strategic interest in raising entry barriers through compliance costs that large incumbents can absorb more easily than competitors. Politico

For policy professionals, the Nvidia position is the more politically potent one in the current U.S. context: Trump's stated view aligns with Huang's, and the administration has shown no appetite for mandatory safety frameworks. Republican legislators including Chip Roy are raising concerns about AI control, but from a governance and human oversight angle rather than a regulatory mandate angle — Roy's concern appears to be about executive branch concentration of AI decision-making, not about imposing compliance requirements on industry. Politico

UK First Secretary Louise Haigh's public warning about AI's 'huge risks' to national security and her call on Prime Minister Andy Burnham to use the 2027 G20 presidency to drive global regulation represents a meaningful shift in UK government positioning — from the previous administration's 'pro-innovation' framing toward a more precautionary, multilateral approach. The UK holds the G20 presidency in 2027, giving it genuine convening power, though the gap between a G20 declaration and enforceable international norms remains vast. The Guardian

Separately, a hybrid side event at UNGA81 High-Level Week hosted by Access Now focused on recentering human rights — particularly surveillance concerns — within the UN AI governance agenda. The UN process remains at the agenda-setting and norm-articulation stage with no binding instrument in prospect, but UNGA81 is a useful indicator of which human rights issues are gaining multilateral salience. The emphasis on surveillance reflects the priorities of civil society organizations from the Global South, where state AI deployment for population monitoring is the primary governance concern rather than frontier model safety. Access Now