Google is calling for AI regulation — but on terms that look remarkably favorable to its existing business practices. In a 21-page policy paper titled “A Pragmatic Approach to AI Governance in America,” released alongside a blog post from Google president Kent Walker, the company proposes what it calls a middle path between over-regulation and the current regulatory vacuum.
“It’s a false choice,” Walker wrote. “There is a middle way: a pragmatic, evidence-based approach that recognizes the unique challenges and opportunities of both frontier AI and widely-deployed AI applications.”
The centerpiece: a new regulator called FARO.
Google proposes a Federally Overseen Frontier AI Regulatory Organisation (FARO), modelled after industry-funded, notionally independent bodies such as the North American Electric Reliability Corporation (NERC) and the Financial Industry Regulatory Authority (FINRA). These organisations are overseen by a government commission but funded and largely shaped by the industries they regulate.
Critics were quick to point out the irony. AI executives have spent years warning that the technology poses existential risks comparable to nuclear weapons or pandemics. Yet the regulatory model Google proposes resembles the kind of corporate self-oversight that has governed Wall Street and the energy sector — bodies that have historically been criticized for regulatory capture.
Copyright: Google’s most self-interested stance.
On the hottest legal question in AI — whether training models on publicly available web data infringes copyright — Google argues it should remain protected as fair use. The company compared training data collection to “an art student taking inspiration from walking through a gallery.”
The Register’s analysis dismantled this analogy: an art student does not control the tourist referral market, capture the entire Louvre’s imagery, sell access to laundered variations, and discourage tourists from visiting the real museum. Several high-profile copyright cases against AI companies are currently winding through US courts, and Google’s position aligns squarely with its own financial interests as a company that scraped vast amounts of web data to train its Gemini models.
Content safety: reasonable measures, but no teeth.
Google’s paper calls for “reasonable measures” on AI platforms: persistent disclaimers, filtering out sexually explicit or romantic content, avoiding claims that a model is a person, and discouraging emotional dependency. But the proposal suggests no binding enforcement mechanism — a pattern critics say echoes Section 230’s failures, where platform safety remained performative rather than enforceable.
Datacenters: the expansion question.
Communities across the political spectrum — including the NAACP, conservative groups, and Atlantic residents — have pushed back against the proliferation of AI data centres, citing water usage, noise, and power grid strain. Google’s position: “The question is not datacenters or no datacenters, but how to build datacenters the right way, responsibly and in partnership with communities.”
The lobbying backdrop.
AI industry lobbying spending has surged 340 percent since 2023, according to an OpenLobby investigation cited by The Register. The implication is clear: the industry is investing heavily to shape whatever regulatory framework emerges. Google’s “middle way” arrives at a moment when Anthropic’s Fable 5 and Mythos 5 models have been suspended under government pressure — exactly the kind of scenario Google presumably wants to avoid for itself.
The paper may be positioned as a reasonable compromise, but between the lines, the message reads differently: let us help write the rules — just make sure they let us keep doing what we are already doing.
Sources: Google wants AI regulation, but on its own terms (The Register, June 2026); A Pragmatic Approach to AI Governance in America (Google policy paper PDF); Kent Walker blog post (Google Blog)