Google president Kent Walker pushes a “middle way” AI regulator model that suits its business

For the last more than three years, AI leaders have been calling for government regulation, then pulling back when oversight could actually cost them business. Now Google is trying to thread that needle with a policy pitch it frames as “a middle way” between over-regulation and no regulation. In a blog post, Google president Kent Walker says the current AI governance debate is stuck in a “false choice,” and that the answer is “a pragmatic, evidence-based approach” aimed at both frontier AI and widely deployed AI applications.

Walker does not put a neat definition around “over-regulation,” but the subtext in the source is unmistakable: Google is presumably responding to the sort of enforcement that can hit specific model releases, including the recent suspension-related controversies around Anthropic’s Fable 5 and Mythos 5. And Google’s policy paper is candid about what it wants that regulation to do. It lays out a “middle path” to “balance market-driven innovation and independent oversight,” with a federally overseen frontier AI regulatory organization called FARO.

FARO is the center of gravity in Google’s argument, and the company tries to make it feel familiar. The paper proposes a regulatory organization modeled after “notionally independent, industry-funded organizations” that still sit under government commission or agency oversight. The source names several comparators: the North American Electric Reliability Corporation, the Financial Industry Regulatory Authority, and the American Medical Association. The logic is simple: you get compliance structures that look independent, but the government still has a thumb on the scale.

That structure matters because it changes what “oversight” means in practice. The source points out a tension between the dire warnings about existential risk from technology leaders and the lived reality of how these systems are allowed to operate online. If AI is truly an existential threat, the source argues, you might expect regulation in the same category as lead or asbestos. Instead, Google is arguing for targeted behavioral requirements that look closer to platform policy than product safety enforcement.

In the paper, Google suggests that AI platforms should take “reasonable measures” including persistent disclaimers, filtering out sexually explicit or romantic content, avoiding claims the model is a person, and regularly stating that it is not. It also argues these requirements should be accompanied by market and innovation incentives through the proposed governance model. The source then drily notes how similar “performative safety measures” have played out on the internet, specifically citing Section 230 of the Communications Decency Act as the legal backbone that can immunize platforms that adopt visible safeguards while misinformation and other harms still spread.

The source’s broader point is that the “middle road” is already here. It lists harms that have become normal in the current ecosystem, including guidance toward harms, model bias, indemnified errors, copyright surrender, and even “chatbot suicide promotion” and “non-consensual nudification images.” The punchline is not that these problems never occur. It is that the regulatory system has so far produced negotiated terms rather than community-wide bans. Google even echoes that framing on infrastructure, saying the question is not “datacenters or no datacenters,” but “how to build datacenters the right way, responsibly and in partnership with communities.”

But if datacenters are politically radioactive in many communities, the source argues that Google’s version of a “middle way” can read like “just let us have our way.” The paper also pushes hard on copyright and training rights. It argues that using publicly available web data for training is a “transformative, non-expressive use” protected under U.S. fair use and text-and-data mining exceptions abroad. The source compares this framing to a tourist-and-art analogy: an “art student” who controls the tourist referral market by capturing the entirety of the Louvre’s imagery and selling access to those images, then discouraging tourists from visiting the actual Louvre, with variations designed to avoid direct copying. Then it adds a final twist: the “art student” claims goodwill by offering incentives for job retraining programs.

For executives watching this, the strategic stakes are bigger than one paper or one regulator model. The source notes AI lobbying has grown dramatically, stating it is up 340 percent since 2023. The subtext for boards and leadership teams is that governance proposals are not just about safety language. They are about who gets to set the boundaries on training data, deployment, and the definition of acceptable risk. And if the industry’s “middle path” keeps steering regulation toward market-driven innovation with independent oversight structures that resemble industry funding models, peers should expect the rules of the game to be written in a way that preserves the ability to scale. The unanswered question is whether that will be enough to satisfy lawmakers and the communities being asked to host the compute.