Federal AI regulation in the United States remains stalled. A proposed White House oversight body is reportedly in limbo, while congressional efforts such as the FRONTIER Act and the AI Kill Switch Act face partisan disagreement, uncertain votes and a shrinking legislative calendar. Neither proposal had become law as of September 16.
The result is an unusual split: Washington has competing ideas for supervising the most powerful AI systems, but no settled federal framework. State governments, meanwhile, continue to enact their own AI and data-center rules.
Federal AI oversight is stuck at the starting line
The White House and the Treasury Department reportedly explored a FINRA-style body for frontier AI. FINRA, the US financial-industry regulator, is a useful analogy here: the proposed model would have connected oversight to an industry that includes the leading AI labs, with those companies playing a role in regulating frontier-model development.
The concept was preliminary rather than an enacted agency or legal regime. It reportedly lost momentum after technology executives opposed the approach and Donald Trump became hostile to overt federal regulation. The idea followed a July 14 essay and briefings by Demis Hassabis, the Google DeepMind cofounder.
By September 16, federal oversight work was reported to be stalled. That left lawmakers with narrower legislative proposals rather than a single comprehensive framework.
What the main proposals would do
The competing approaches focus on different points of control: transparency and audits for advanced developers, or an emergency mechanism for stopping especially powerful systems.
| Proposal or framework | Scope | Main mechanism | Relevant date |
| White House FINRA-style framework | Frontier AI developers | Proposed administration-connected oversight body involving self-regulation by leading AI labs | Preliminary concept reported in September 2026 |
| FRONTIER Act | Largest developers and most advanced models | Model cards, risk-management frameworks, independent audits, incident reporting and ongoing assessments | Introduced July 23, 2026 |
| AI Kill Switch Act | Certain powerful AI systems capable of catastrophic harm | Technical ability to throttle, suspend or fully shut down covered systems, plus incident reporting and forensic records | Introduced July 23, 2026 |
The FRONTIER Act was introduced by Lori Trahan and Jay Obernolte with bipartisan support. Its stated design is a national, risk-based framework with requirements that increase for larger developers and more advanced models. The proposal would make developers document their systems, manage identified risks, undergo independent evaluation and report serious incidents.
A separate description of the bill included the possibility of Commerce Department auditors being embedded at AI labs and of development being halted after findings involving catastrophic risk. Those provisions remain part of the reported legislative concept rather than a federal rule.
The AI Kill Switch Act, introduced by Ted Lieu and Nathaniel Moran, takes a more direct approach. Covered developers would have to retain the technical capacity to throttle, suspend or shut down certain systems. The proposal also includes graduated government responses, incident reporting and forensic records intended to preserve a trail of what happened.
That is why “kill switch” can be misleadingly simple. A shutdown capability addresses whether a system can be stopped or restricted after intervention is triggered. Some technology executives have argued that harmful behavior could occur while a system is unobserved or driven to complete a task, meaning a shutdown after discovery might not prevent damage that had already occurred.
Why Washington is not moving
The federal deadlock has several overlapping causes. The White House has resisted overt regulation, technology executives opposed the proposed oversight framework, and lawmakers disagree over how much authority the government should have over advanced AI developers.
The national-security argument is central. Critics of strict limits warn that slowing US AI development could give China an advantage. Supporters of stronger safeguards argue that rapid progress without enforceable controls creates its own security risks. That disagreement cuts across the policy debate and makes a broad compromise harder to assemble.
Congressional procedure adds another obstacle. More than 100 AI-related bills have been introduced over two years, but the House had one week in session and the Senate three weeks before the November midterm elections, according to a September 14 account of the legislative calendar. House Democratic leaders reportedly considered attaching AI audit provisions to must-pass government-funding legislation, but the available votes and Speaker Mike Johnson’s support remained uncertain.
The FRONTIER Act and the AI Kill Switch Act therefore entered Congress as concrete proposals, but they faced a difficult path through committees, leadership and floor votes. As of September 16, neither had become law.
Federal paralysis does not mean no US AI policy
The federal standstill has not stopped state-level activity. A tally published for July 1, 2026, counted 109 enacted state AI laws and 28 data-center laws. State measures have addressed areas including child safety, companion chatbots, consumer protection, data centers and selected rules for advanced models.
California, New York and Illinois were highlighted as states taking more substantive action while federal efforts remained stalled. That creates a patchwork for developers and users: the rules can differ depending on where an AI company operates, where infrastructure is built and which people or services a state law covers.
The immediate US policy landscape is therefore not an absence of ideas. It is a pileup of partially competing ones: a reported executive-branch oversight concept in limbo, a transparency-and-audit bill, shutdown requirements for powerful systems and active state legislation. The next federal step depends on whether lawmakers can turn any of those proposals into a measure with enough support before the congressional calendar closes.
