A Regulator With Teeth: Welch and Bennet Unveil the AI Regulator Act

Two Democratic senators have just put forward the most aggressive federal AI oversight framework Washington has seen this year. On Wednesday, September 23, Senator Peter Welch of Vermont — a member of the Senate Judiciary Subcommittee on Antitrust, Competition Policy, and Consumer Rights — joined Senator Michael Bennet of Colorado in announcing the AI Regulator Act, a proposal that would empower a new independent federal agency to pre-certify frontier AI models, halt risky releases, and impose some of the largest civil penalties ever contemplated for a technology sector.

What the bill would actually do

The AI Regulator Act builds on the senators’ earlier Digital Platform Commission Act, first introduced in 2022 and reintroduced in 2023, which sought to create a dedicated regulator for the largest digital platforms. The new proposal supercharges that vision with a set of authorities aimed squarely at frontier AI development:

Pre-certification for frontier models. Developers of the most capable AI systems would need to clear regulatory review before public release — a structure closer to pharmaceutical or aviation approval than to anything the software industry has faced before.

The power to pause. The proposed Federal Digital Commission could delay the public distribution of any AI model judged to carry catastrophic-risk potential for up to six months, or until reasonable safeguards are in place. That is a genuinely striking provision: no U.S. regulator today can unilaterally stop a model launch.

Penalties that bite. The five-member commission could hold hearings, launch investigations, conduct research, issue binding rules, and impose civil penalties of up to 15 percent of a firm’s prior-year global revenue. For a company with $50 billion in annual revenue, that is a theoretical $7.5 billion exposure — an order of magnitude beyond typical FTC fines.

Systemic-importance designations. Echoing the framework regulators use for big banks, the commission could designate “systemically important digital platforms or developers” for heightened reporting and supervision.

Frontier safeguards. Developers would face mandatory risk-mitigation obligations, catastrophic-risk incident reporting, and transparency requirements.

Online safety and national security provisions. The package also includes age-verification and age-appropriate design standards, consumer protections, transparency rules for terms of service and content moderation, and — notably — provisions governing Americans’ data held by foreign AI companies, U.S. subsidiary requirements for certain foreign firms, and disclosure of where online content originates.

The rhetoric is escalating too

The senators’ framing was unusually blunt for a legislative rollout. “Americans in Vermont and across the country are freaked out by the warnings of AI researchers, and I’m right there with them,” Welch said. “It’s clear we can’t leave it to AI companies to self-regulate — just like we don’t let drug companies, or Wall Street, or Big Oil self-regulate.”

Bennet reached for historical precedent: the U.S. created expert agencies for aviation, pharmaceuticals, and telecommunications when those industries transformed the economy, he argued, but “today, no such agency exists for AI or social media platforms,” leaving oversight fragmented across states and federal agencies while Congress “tries to address each new problem one bill at a time.”

The timing is not accidental. The announcement lands amid a remarkable week of AI-governance news: OpenAI, Anthropic, and Google DeepMind have reportedly been in weeks-long talks about coordinating on safety; two frontier-lab founders addressed the UN Security Council on loss-of-control risks; and just yesterday the U.S. and China neared agreement on an AI crisis hotline. Even OpenAI itself published an essay on September 9 titled “The AI policy window is open,” calling for mandatory national AI safety requirements. When the industry’s most valuable company is asking Congress to regulate it, the political cover for ambitious proposals grows.

Why this faces long odds

The AI Regulator Act is a proposal, not a law — and the procedural path from here is steep. The U.S. Congress has not passed a comprehensive AI statute despite more than 150 AI-related bills introduced in recent sessions. The current administration’s posture cuts against this bill’s direction: a December 2025 executive order on “Eliminating State Law Obstruction of National Artificial Intelligence Policy” moved to discourage state-level AI regulation and push toward a lighter, innovation-first national framework, and congressional Republicans have repeatedly sought moratoriums on state AI enforcement — a ten-year moratorium was stripped from a budget package by a 99–1 Senate vote in July 2025, and preemption fights have continued since.

There is also intra-party complexity on the Democratic side. Just days ago, Representative Mike Lawler, a New York Republican, introduced the “Stop Rogue AI Act” while explicitly opposing any pause in AI development — evidence that even lawmakers alarmed about AI misuse reject the halt-based approach at this bill’s core.

And the industry is split. Some frontier labs now publicly welcome federal safety standards, partly because a patchwork of state laws — California, Texas, Illinois, and Connecticut all enacted significant AI legislation taking effect in 2026 — is harder to comply with than a single national regime. But pre-certification and six-month pause authority would represent a step beyond anything the major labs have endorsed, and the 15%-of-global-revenue penalty is modeled on EU-style deterrence rather than American administrative practice.

What to watch

Three signals will tell you whether the AI Regulator Act is a serious vehicle or a positioning exercise. First, whether it picks up any Republican co-sponsors — bipartisan cover is the necessary condition for any AI bill to move. Second, whether the major labs lobby for, against, or silently tolerate it; a public endorsement from even one frontier developer would mark a watershed in the industry’s Washington strategy. Third, whether the proposal gets folded into the broader preemption debate — a federal commission with real enforcement power could be traded to states as the price of displacing their own AI laws.

Either way, the Overton window has moved. A six-month government pause on a frontier model and nine-figure-percentage fines were fringe ideas in 2023; in September 2026, they are in a Senate press release with two senators’ names on it. The gap between American ambition and American legislative reality remains wide — but proposals like this one define the terms of the debate that eventual law will be written in.