One Republican Had a Plan to Tackle AI. Why Did House GOP Leaders Say No — and What Happens Now?
By Saqib S. Qureshi — SCN News
WASHINGTON, Sept. 18 — Nearly two years ago, a bipartisan group of U.S. lawmakers produced one of Congress’s most comprehensive attempts to answer a rapidly approaching question: how should Washington govern artificial intelligence without crippling the technology it wants American companies to lead? The House task force spent months consulting industry executives, researchers, government officials and civil-society experts before producing 66 findings and 85 recommendations covering everything from national security and privacy to jobs, health care, intellectual property and the risks posed by increasingly powerful AI systems.
For Republican Rep. Jay Obernolte of California, who chaired the bipartisan task force alongside Democratic Rep. Ted Lieu, the report was supposed to be a beginning rather than an endpoint. Obernolte wanted House leaders to create a dedicated congressional committee capable of turning the sprawling recommendations into legislation and navigating the overlapping jurisdictions that make technology policy notoriously difficult on Capitol Hill. Republican leaders declined. A senior GOP leadership aide told CNN that the leadership believed the original task force had completed its assignment and existing congressional committees could take the recommendations from there.
Almost two years later, that decision has become part of a much larger argument over whether Congress can move quickly enough to govern a technology evolving faster than its legislative machinery. Lawmakers have introduced bipartisan proposals addressing catastrophic AI risks, independent audits, incident reporting and even emergency shutdown mechanisms. But Congress has yet to establish a comprehensive federal system governing the most powerful frontier models, while President Donald Trump and House Speaker Mike Johnson have warned against regulations they believe could slow American innovation and give China a strategic advantage.
The result is not simply a partisan dispute over whether AI is dangerous. Republicans and Democrats have produced bipartisan legislation acknowledging significant risks. The deeper problem is that Washington still disagrees about who should regulate AI, which risks require mandatory rules, how much authority federal agencies should receive, whether states should be allowed to establish their own standards, and how far safeguards can go before they begin slowing American development of a technology increasingly treated as an economic and national-security competition with China.
That unresolved conflict is becoming more urgent as leading AI executives themselves call for government oversight. Anthropic’s Dario Amodei, OpenAI’s Sam Altman and xAI’s Elon Musk have all recently warned about advanced AI risks, while lawmakers from both parties have proposed new safeguards. Yet Johnson said this week that AI companies can self-regulate and rejected a moratorium on development, arguing that slowing American systems would threaten the U.S. competitive position against China.
What exactly did Obernolte want Congress to do?
The origins of the current impasse go back to February 2024, when Johnson and House Democratic leader Hakeem Jeffries created a 24-member bipartisan Task Force on Artificial Intelligence. Twelve Republicans and 12 Democrats were appointed, with Obernolte as chairman and Lieu as co-chairman.
That structure mattered because AI does not fit neatly inside a single congressional committee. One AI system can simultaneously raise questions involving commerce, national security, cybersecurity, copyright, employment, financial regulation, health care, privacy, education and energy. Each subject can fall under a different committee, creating jurisdictional disputes before lawmakers even reach the substance of legislation.
The task force attempted to overcome that fragmentation by building a shared congressional framework. Its final report described AI as potentially transformative for productivity, medicine and national competitiveness while also acknowledging misuse and emerging harms. The document set out 66 findings and 85 recommendations and called for a targeted approach that would encourage innovation while creating appropriate safeguards.
Obernolte subsequently wanted a special committee that could continue the work and draft legislation. Instead, Republican leadership concluded that standing committees could process the recommendations themselves. That decision did not prohibit Congress from legislating on AI, but it removed the centralized mechanism Obernolte believed could coordinate the issue across committee boundaries.
“I think the need for action is urgent,” Obernolte told CNN, saying lawmakers had been making that argument for roughly a year and a half.
The consequences can now be seen in the congressional pipeline. The House Science, Space and Technology Committee advanced 10 AI-related measures in June, but they had not reached the House floor by this week. Other proposals are spread across multiple committees, while lawmakers openly acknowledge that Congress lacks a single institutional center responsible for developing AI policy.
Congress has proposals. The problem is getting them into law.
One of the most important current attempts is the FRONTIER Act, introduced in July by Obernolte and Democratic Rep. Lori Trahan with additional Republican and Democratic co-sponsors. Its full name — the Frontier Risk Oversight, National Transparency, Independent Evaluation, and Reporting Act — reflects what its authors are trying to create: a federal system specifically aimed at the most advanced AI models rather than blanket regulation of every AI product.
The legislation proposes tiered requirements depending on the size of the frontier-model developer. Those requirements would include model cards explaining systems, formal risk-management frameworks, independent audits, incident reporting and continuing assessments of potentially catastrophic risks. Its sponsors argue that a single national framework could simultaneously create safeguards and prevent companies from confronting a different regulatory regime in every state.
That is significantly different from banning AI development or ordering companies to stop training advanced systems.
Yet even this bipartisan approach is struggling to move.
House Energy and Commerce Committee Chairman Brett Guthrie said this week that he would not promise a committee vote on the FRONTIER Act and suggested substantive action could slip into 2027. Obernolte wants committee consideration in November, but Guthrie said the legislation is complicated and that he did not want Congress rushing such a consequential issue during a lame-duck session.
The House calendar makes the problem harder. With lawmakers leaving Washington for the election campaign and only limited legislative time remaining before the November midterms, even proposals enjoying support from members of both parties face a procedural bottleneck. Senate Majority Leader John Thune has separately discussed a path forward with Democratic Sen. Amy Klobuchar on legislation addressing catastrophic risks from frontier models, but the two chambers have not converged on a comprehensive package.
Another bipartisan proposal would require developers of the most powerful AI models to maintain a mechanism capable of shutting them down under specified emergency circumstances. The proposal reflects growing concern about highly autonomous systems causing catastrophic harm, but it too illustrates the fragmented congressional approach: lawmakers are pursuing individual pieces of AI governance while a comprehensive architecture remains unsettled.
Why are Republican leaders reluctant?
The disagreement is partly philosophical and partly geopolitical.
Johnson has made clear that he does not support a moratorium on AI development. He argues that the United States is competing directly with China and that stopping or substantially slowing American companies could create national-security consequences if Chinese systems continue advancing. On Tuesday, he said AI companies could regulate themselves and did not need the government ordering them to slow development.
That does not mean Johnson has rejected every safeguard. He has spoken favorably about independent auditing and greater transparency from developers and has called for leading AI executives to meet with Trump at the White House. His position is therefore better characterized as resistance to government-led restrictions on development rather than rejection of every form of AI oversight.
The Trump administration has taken a similar competitiveness-first approach. A White House executive order says U.S. AI companies should be able to innovate without “cumbersome regulation” and argues that inconsistent state laws could create a costly patchwork for developers, particularly smaller companies. The administration has directed federal officials to pursue a minimally burdensome national framework and challenge state rules it considers inconsistent with that policy.
That creates a complicated distinction inside the Republican debate. Obernolte also emphasizes American technological leadership and competition with China. He is not proposing that the United States stop developing AI. In congressional remarks earlier this year, he said a federal framework should preserve American leadership while confronting cyberattacks, privacy threats and other emerging risks.
The argument is therefore not simply regulation versus no regulation.
It is a dispute over what kind of regulation can reduce serious risks without slowing the companies Washington believes are competing with China for control of a strategically important technology.
Why does Washington suddenly seem more worried?
The political environment surrounding AI has changed rapidly.
For several years, congressional debates focused heavily on familiar problems: deepfakes, privacy, discrimination, copyright, misinformation, job displacement and children’s safety. Those issues remain unresolved. But increasingly capable frontier models have added another category — systems capable of autonomously performing complex tasks, discovering software vulnerabilities, conducting cyber operations or potentially helping users acquire dangerous scientific capabilities.
Recent warnings from AI industry leaders have pushed those concerns further into mainstream politics. Amodei, Altman and Musk have called for stronger oversight, creating the unusual situation in which executives developing the technology are themselves asking government to establish safeguards.
That does not establish that catastrophic AI outcomes are imminent.
The distinction is essential. Concerns about future loss of control or existential harm involve uncertain projections about systems considerably more capable than most AI deployed today. Other harms — fraud, deepfakes, cyber misuse, privacy violations and disruption to employment — already exist or can be observed more directly.
Congress is being asked to write legislation that addresses both categories without treating speculative worst-case scenarios as established facts.
That is extraordinarily difficult because regulation written around today’s systems could become obsolete quickly, while rules designed around hypothetical future capabilities risk regulating technology that does not yet exist in the form lawmakers anticipate.
Then why not let the states handle it?
Because state regulation has become another major front in the battle.
In the absence of comprehensive federal AI legislation, states have increasingly developed their own rules. Supporters say states cannot simply wait for Washington while AI systems are deployed in employment, health care, education and other high-impact settings.
The Trump administration sees a different danger: 50 potentially inconsistent regulatory systems governing companies whose models operate nationally. Its executive order directs the administration to identify and challenge state AI laws it regards as excessively burdensome and calls for Congress to create a uniform federal framework capable of pre-empting conflicting state rules.
Democratic lawmakers have pushed back. Reps. Don Beyer, Doris Matsui, Ted Lieu, Sara Jacobs and April McClain Delaney introduced legislation in March aimed at reversing the administration’s effort to restrict state AI regulation. They argue that Washington should not prevent states from adopting safeguards while Congress has failed to establish equivalent federal protections.
The contradiction is becoming increasingly difficult to avoid: Washington is debating whether states should be prevented from regulating AI before Washington itself has agreed on comprehensive federal rules to replace them.
Is Congress actually doing nothing?
No. That would overstate the situation.
Congress has enacted narrower technology protections and continues moving legislation affecting AI-related infrastructure. Trump signed legislation targeting non-consensual explicit images, including AI-generated material, and the House this week overwhelmingly approved legislation aimed at preventing electricity customers from absorbing higher utility costs associated with rapidly expanding data centers.
Committees are also working on research, standards, cybersecurity, transparency and other targeted AI issues.
But those actions are different from establishing federal rules governing frontier AI development itself.
That distinction explains why critics say Congress has failed to regulate AI despite years of hearings, task forces and legislative proposals. The institution has demonstrated that it can address discrete harms. What it has not yet demonstrated is that it can agree on the broader architecture: who supervises frontier developers, what information companies must disclose, when independent testing becomes mandatory, what qualifies as an unacceptable risk and what authority government should possess when a model crosses that threshold.
Could industry self-regulation work?
That is now one of the central unresolved questions.
Johnson's position is that companies have strong incentives to control dangerous systems and can voluntarily slow development if they believe their models are unsafe. Independent audits and transparency requirements could reinforce that process without requiring a large federal regulatory apparatus.
Supporters of mandatory rules question whether market incentives are sufficient when companies are racing one another — and foreign competitors — to release increasingly capable models. A company that voluntarily delays a powerful system for extensive safety testing could lose customers, investment or technological advantage to a competitor that moves faster.
The FRONTIER Act attempts to bridge those positions. Instead of imposing a blanket development moratorium, it would require increasingly significant safeguards as developers and models cross defined thresholds. Independent auditing and incident reporting would make at least part of the safety process externally verifiable rather than leaving companies entirely responsible for assessing themselves.
Whether Congress ultimately accepts that model remains uncertain.
What happens next?
In the short term, comprehensive House action appears difficult.
Guthrie has declined to commit to advancing the FRONTIER Act this year, while the House calendar is shrinking ahead of the November election. Johnson has called for discussions with AI executives rather than Congress taking the immediate lead in imposing broad restrictions. The Senate, meanwhile, continues exploring bipartisan legislation around catastrophic-risk safeguards.
That leaves Washington in an unusual position.
The House already completed a major bipartisan study. Its members agreed on dozens of findings and recommendations. A Republican who chaired that effort wanted Congress to create a dedicated structure for converting the work into legislation. Republican leadership declined, believing existing committees could do the job. Those committees now have multiple bills before them, including a bipartisan frontier-model framework, but the most consequential measures remain unfinished.
At the same time, some of the world's most prominent AI executives are asking Washington for stronger oversight, lawmakers from both parties acknowledge potentially serious risks, and the White House argues that excessive regulation could jeopardize America's technological advantage over China.
The unresolved issue is therefore larger than Obernolte's rejected committee proposal.
It is whether the U.S. political system can build rules for a technology while that technology is changing faster than the process used to write those rules — and whether Washington can distinguish between safeguards that reduce genuine risks and restrictions that could unintentionally weaken innovation.
Congress has already spent years studying that question.
It has not yet agreed on the answer.