US AI Regulation Stalls as White House Shelves Oversight Plans and Bills Stall in Congress

Federal efforts to regulate artificial intelligence have ground to a halt in Washington, constrained by White House resistance and congressional gridlock despite growing public debate over frontier model safety. A proposed administrative framework for a FINRA-style self-regulatory body has been shelved following pushback from major tech leaders, while competing legislative proposals on Capitol Hill lack the requisite support to advance before upcoming elections, according to reporting by Wired.

White House officials had initially begun outlining an administrative oversight body after Google DeepMind cofounder Demis Hassabis suggested a self-regulatory framework in a July 14 essay and in briefings with senior leaders. The Treasury Department and the White House Office of Science and Technology Policy drafted preliminary plans that would allow leading artificial intelligence laboratories to oversee frontier model development. However, the initiative hit a standstill in August after high-profile technology executives—including Meta Chief Executive Mark Zuckerberg and former administration AI adviser David Sacks—contacted President Donald Trump directly to express opposition.

Following those interventions, the White House paused active work on national AI oversight measures. Trump has repeatedly rejected formal regulation, recently publishing a post on Truth Social that characterized AI safety concerns as a "hoax." According to administration sources, Trump's stance was further hardened by a 3,822-word treatise authored by Anthropic Chief Executive Dario Amodei regarding the systemic risks posed by frontier AI. White House aides viewed the essay as creating unnecessary political friction, despite representing a moderate position within the broader technology sector.

In the absence of executive action, legislative efforts in Congress face severe procedural and political hurdles. With no remaining voting days scheduled on the House calendar prior to the midterm elections, the likelihood of enacting meaningful AI oversight remains low. The most widely supported proposal, the Frontier Act, was introduced earlier this summer by Representative Lori Trahan, a senior House Democratic whip, and Representative Jay Obernolte, a Republican member of the House Energy and Commerce Committee.

The Frontier Act would grant the Department of Commerce authority to embed auditors inside frontier AI laboratories to track critical safety events and empower administration officials to pause development if a model creates catastrophic risks. While OpenAI and the House Democratic caucus have backed the measure, the White House has withheld endorsement. Internal skepticism stems from industry warnings regarding regulatory capture as well as national security concerns surrounding China. Critics contend that forcing American labs to halt development during safety pauses would allow Chinese entities to continue training and reverse-engineering frontier AI weights without interruption.

Legislative momentum for the Frontier Act is further impeded by partisan leadership dynamics. House Speaker Mike Johnson has signaled no intention to bring the bill to the floor for a vote, leading House Democratic aides to evaluate incorporating key auditing provisions into must-pass funding measures, such as a federal continuing resolution. Furthermore, Democratic support is not unanimous; Representative Josh Gottheimer, who served on the House Democrats' AI commission, has declined to sign onto the legislation.

A secondary legislative track involving "kill-switch" mechanisms—authored separately by Representative Ted Lieu, vice chair of the House Democratic caucus, and Louisiana Republican Senator John Kennedy—would give federal authorities the power to forcibly shut down rogue models. However, Silicon Valley executives have shown little enthusiasm for the concept. Industry leaders privately noted that severe safety failures are rarely observed in real time during testing, citing incidents such as the Hugging Face breach, which was discovered only after completion, rendering post-hoc kill switches largely ineffective.