The White House has paused work on creating a FINRA-style regulatory body for artificial intelligence after President Donald Trump soured on the idea, and no new AI oversight proposals are seriously under consideration there. The measure had been sketched out by the Treasury Department and the Office of Science and Technology Policy. Its collapse leaves the United States without a federal mechanism for supervising frontier models, even as the debate over their risks dominates public discussion.
How the oversight agency was shelved
White House aides began drafting a framework for an AI oversight agency soon after Google DeepMind cofounder Demis Hassabis floated the idea in a July 14 essay and in briefings to senior officials, according to sources familiar with the matter. The Treasury Department and the Office of Science and Technology Policy envisioned a body in which leading AI labs would self-regulate the development of frontier models, and they prepared a preliminary draft to see what execution would look like. In August, a group of tech executives, including Trump's former AI czar David Sacks and Meta CEO Mark Zuckerberg, learned of the effort and called the president one by one to express opposition. The proposal has since been left in limbo.
The mechanics of the abandoned plan rested on self-regulation rather than state inspection: laboratories would police their own frontier development, with the government providing the framework rather than day-to-day supervision. That design differed from the audit model now discussed in Congress, where independent auditors would sit inside AI labs. The White House has been undecided on whether to support independent auditors at all, facing skepticism from AI accelerationists concerned about regulatory capture and from national security hawks focused on China. Trump has also hardened his position in recent days, rejecting overt regulation and calling AI safety concerns a hoax in a Truth Social post.
The retreat follows a broader pattern in Washington, where the appetite for binding rules has narrowed over the past months. Trump's irritation, according to one source, was stirred at least in part by a 3,822-word missive from Anthropic CEO Dario Amodei on the dangers of frontier AI, which aides viewed as amplifying anxieties and creating unnecessary headaches for the administration, even though the position is fairly moderate within AI circles. In the absence of White House leadership, several bills have appeared in Congress, but with no voting days left on the House calendar until after the midterm elections and no unity around a single bill, the odds of passage are extraordinarily slim.
What this means for companies deploying AI
For businesses building on frontier models, the practical consequence is that no federal licensing or audit requirement is coming in the near term, and vendor selection will continue to rest on internal risk assessment rather than a government seal. The bipartisan Frontier Act, introduced this summer by House Democratic senior whip Lori Trahan and Republican House Energy and Commerce Committee member Jay Obernolte, is widely seen as having the most support inside the House Democratic caucus and among AI executives. As drafted, it would let the Commerce Department embed auditors at AI labs, who could report critical safety incidents and empower the administration to stop development if a model poses catastrophic risk. OpenAI has publicly backed the bill, but the White House remains undecided on the auditor concept.
Even if the Frontier Act gathered enough votes, Republican Speaker Mike Johnson has no interest in bringing it to the floor, according to House Democratic leadership aides. To get around the procedural hurdles, Democrats are looking at attaching some of the bill's AI audit provisions to must-pass legislation such as a continuing resolution to fund the government, according to two House leadership aides. Unified support is not guaranteed even within the Democratic caucus: some key lawmakers, including New Jersey congressman Josh Gottheimer, who served on the House Democrats' AI commission, have not signed on. A separate kill-switch idea, which would let the government force a model to shut down, has bipartisan and bicameral backing but only modest interest from Silicon Valley executives.
The main argument against auditors in Trumpworld is that even if a model is judged dangerous and taken offline, China is not guaranteed to pause its own development or stop trying to reverse engineer frontier AI weights. That would create immense pressure to lift any emergency pause to stay ahead of China, defeating the premise of having auditors push the Commerce Department to take models offline. Tech leaders have also been largely ambivalent about kill-switch bills, since the primary concern of AI engineers is not how models behave when observed but how they behave when they believe they are unobserved and incentivized to take extreme steps to complete a task. The Hugging Face hack, for instance, was discovered only after it was over, which some executives privately told WIRED suggests little point in switching a model off after the damage is done.
The marker to watch is whether the audit provisions of the Frontier Act are attached to a must-pass spending measure, since that route would bypass the Speaker's control of the floor and give companies their first concrete signal that federal oversight of AI labs is moving from proposal to law. Until then, corporate AI programs should assume no external safety gate will be imposed, and should document their own testing and incident reporting as the only record available to regulators, customers and insurers.
