• Proposes model to oversee frontier AI safety compliance
• Urges grid buildout to support data center growth

Google has released a white paper arguing for a new federal framework for artificial intelligence governance, pairing a proposed regulatory path for frontier model safety with a call for faster US energy and transmission buildout to support data centers.

In the paper, A Pragmatic Approach to AI Governance in America, Google says the policy debate has been trapped in a false choice between over-regulation and no regulation and instead proposes separate regimes for frontier AI and more widely deployed consumer and enterprise applications.

For frontier models, Google proposes a federally overseen frontier AI regulatory organization, or FARO, modeled on industry-funded bodies operating under government supervision, such as the North American Energy Reliability Corp. and FINRA.

The company says such an entity could move faster than traditional bureaucracy, set national standards for frontier model safety and security, and verify compliance before the most advanced models are released to the US market. Google says oversight could sit with agencies such as the Commerce, Treasury or Energy departments.

The paper says the FARO should develop capability-based benchmarks for frontier models, especially around cyber and chemical, biological, radiological, and nuclear risk, and maintain a repository of standards for how advanced systems are built, tested and deployed.

In the near term, Google says frontier labs should publish and follow their own safety frameworks and undergo annual independent procedural audits, with confidential reports submitted to the FARO. Over time, those audits could evolve into substantive compliance reviews as standards mature.

‘Highway program’

On energy infrastructure, Google argues that AI growth requires what it calls an “Eisenhower Highway Program” for the US grid. The company says lawmakers should pursue comprehensive federal permitting reform to speed energy generation and transmission projects, and it contends that properly structured data center growth can lower electricity costs by spreading fixed grid costs over a larger volume of demand.

The paper also says large data centers should add energy capacity, pay for associated infrastructure, and include safeguards so costs are not shifted to existing customers. Google points to mechanisms such as the Capacity Commitment Framework, the Clean Transition Tariff, and the Ratepayer Protection Pledge as models it supports writing into federal law.

Beyond frontier safety and energy, the paper lays out additional policy asking for widely deployed AI. Google calls for workforce data collection, skilling initiatives, and possible modernization of unemployment insurance if AI-driven displacement rises. It urges child-safety rules requiring chatbot disclaimers, bans on gamified engagement and “pause-and-direct” protocols for self-harm queries. It also backs watermarking and provenance standards such as SynthID and C2PA, continued fair-use treatment for AI training on public web data, and new technical standards plus privacy-enhancing technologies to protect user data.