The search giant's 21 page white paper proposes a federally overseen, industry funded compliance body for the most capable AI systems, and critics say the architecture would lock smaller labs and open source developers out.
Google published a 21-page white paper on Thursday arguing for a federally overseen, industry-funded body to set the rules for the most capable AI systems. The body, which Google calls the Frontier AI Regulatory Organization, or FARO, would be modeled on the regulators that already govern the U.S. power grid, brokerage industry, and medical profession.
The proposal is being marketed as a "middle way" between a comprehensive federal AI law and no rules at all. The more useful question is who would actually bear the cost of compliance once the rules exist. In every regulated industry Google's chosen models were built to resemble (NERC for the power grid, FINRA for brokerages, the AMA for medicine), the entity that ends up writing the rulebook is the same entity that has the lawyers and engineers to keep up with it. Smaller players pay the same dues and often drop out.
The white paper, titled "A Pragmatic Approach to AI Governance in America" and signed by Google president Kent Walker, argues the model would let "expert agencies" write technical rules faster than Congress can and adapt them as the technology changes. Google published the paper alongside a blog post on its public policy page on June 26, with the full 21-page white paper PDF hosted on Google's own servers.
Google's framing is that industry-funded, independent standards bodies keep pace with technical change in ways Congress cannot. Critics say the mechanism that makes that true is also the mechanism that entrenches incumbents. If the proposal genuinely lowered the barrier for a graduate student fine-tuning a small model, or for an open-source collective releasing weights on a public hub, the incumbency critique would collapse. The white paper does not make that case.
The Register first surfaced the proposal and framed it as "Google wants AI regulation, but on its own terms." Several of Google's frontier-lab competitors have taken public positions on the regulation question. Anthropic CEO Dario Amodei called for "binding regulations" earlier in June, then watched his company's Fable 5 and Mythos 5 models get suspended under a U.S. export-control directive on June 13, an order Bloomberg first reported, Anthropic disclosed publicly in a company statement, and Forbes laid out in terms of what changed for users. Google's white paper does not address export controls, but it does argue that "binding" rules should be set by a U.S.-domiciled body.
On copyright and training data, the white paper takes a sharper position. Web-scraped training data should remain "transformative, non-expressive use," the paper argues, and therefore fair use and protected under text-and-data-mining rules abroad. The Register called that framing "Google's biggest ask." It is also the position Google is currently litigating in U.S. courts, where the fair-use question for training data is not settled.
On datacenters, where community opposition has mounted across the political spectrum (civil rights groups, local officials, and national outlets including Axios and The Atlantic have all reported pushback), the paper reframes the fight as "how to build datacenters the right way, responsibly and in partnership with communities." That language directly contradicts the bans many of those communities are pushing for, and treats the question of whether to build at all as already settled.
The proposal lands in a Washington that has heard versions of this argument before. AI industry lobbying has grown roughly 340% since 2023, according to OpenLobby, an advocacy transparency site. Google itself called for AI regulation in Congressional testimony last year, and the New York Times has documented an "AI lobbying blitz" across Silicon Valley.
On the rules Google thinks AI platforms should adopt directly, the white paper is more restrained. It proposes "reasonable measures" (persistent disclaimers identifying AI-generated content, content filtering, avoiding personhood claims, not designing for emotional dependency) and frames these as self-policy rather than hard rules. Whether the new compliance body would convert those soft commitments into enforceable standards is one of the open questions left unanswered in the proposal.
What is still open: FARO has no bill number, no co-sponsors, and no federal home. The paper acknowledges it is a proposal, not legislation. The question for the next year is whether Congress, an agency, or an industry coalition picks up the framework, and whether the compliance architecture that emerges is the one Google's white paper described, or one shaped more visibly by smaller players, open-source developers, and the communities now pushing back on the physical infrastructure that makes frontier AI possible.