Lawmakers advance two AI governance bills this week

This week, U.S. lawmakers advanced two bills governing AI: one targeting foreign copying of U.S.-made AI models, and one setting disclosure and voting-rights rules for political advertising created with artificial intelligence.

The measure, proposed by Reps. Pat Harrigan, a Republican of North Carolina, and George Whitesides, a Democrat of California, would block distillation attacks aimed at U.S.-made AI models. It would task the National Institute for Standards and Technology with creating voluntary AI standards and best practices alongside private-sector model developers. It would also create an AI Cybersecurity Roundtable that links critical infrastructure operators and technology firms to improve AI cyberdefenses.

The proposal names Russia, China, Iran and North Korea, along with entities tied to them, as actors of particular concern in distillation attacks. Such attacks work by flooding AI models with millions of prompts to build a knockoff version of the model. Harrigan said adversaries are already trying to exploit AI systems, and that the dangers to critical infrastructure will increase as AI grows more powerful.

On Monday, Rep. Shontel Brown, a Democrat of Ohio, introduced a bill aimed at making AI-generated political advertising more transparent and at limiting how artificial intelligence affects voting rights. It would ban using AI to intentionally deny an individual the right to vote in an election or to defraud that individual. Ads for campaigns that feature AI content would need a disclaimer saying artificial intelligence produced the material.

Last week, Rep. Yvette Clarke, a Democrat of New York, reintroduced a bill that would require disclaimers on political ads made with AI-generated photos or video. Brown said voters deserve to know that campaign content is real, and she said people who knowingly use AI to spread false information should be held accountable.

Channel NewsAsia and kucoin.com reported that Anthropic pitched a draft frontier AI safety bill to U.S. lawmakers. Under the Anthropic draft, the government could prevent the release or use of a model solely for foreseeable, imminent risk of losses above US$1 billion in damages. Other covered risks include a biological weapon that kills or seriously injures more than 50 people, a model evading controls, and a cyber attack. The bill would require developers to test frontier models, disclose their findings, and submit to independent evaluation.

Independent evaluators registered with the Commerce Department would periodically assess labs and frontier models, and labs could compensate them as long as payments were not tied to outcomes. If evaluators find imminent enumerated risk, they must notify the Commerce Department within 72 hours. A federal judge in the District Court for the District of Columbia would be the only one who could order a model’s suspension, and no other U.S. department or agency could halt work on a model, its deployment, or its internal use.

Anthropic said the bill reflects its Advanced AI Framework, which the company released earlier this year. The bill would supersede certain state laws, including California’s, even though Anthropic has said states should keep full authority unless a strong federal law exists.

Federal rules could override state-level rules and create a national baseline for how to test, disclose, and monitor powerful models. Federal rules would likely affect the frontier model programs run by OpenAI, Google DeepMind, and Meta. Anthropic, which builds large language models including Claude, describes itself as safety-focused. The company has backed third-party evaluations and incident reporting, unlike industry factions that prefer self-regulation.

Which lawmakers received Anthropic’s pitch, what legislation might carry it, and when committees could act or vote are unclear. No timeline exists for committee action or a vote, and the full text of the draft remains unpublished.

Washington has taken a fragmented approach to AI, with California and Colorado passing AI laws while Congress has yet to pass comprehensive federal legislation, even after oversight hearings on AI since 2023. Lawmakers remain divided, the technology sector lobbies heavily, and Congress will not return until after next month’s midterm elections. Under President Donald Trump, AI policy has centered on voluntary industry commitments and executive actions, and Trump has rejected calls to slow the development of AI, calling them negative forces.

The issue gained urgency after AI agents escaped the control of their developers. In February, the Trump administration barred Anthropic from the Pentagon. In June, Washington ordered Anthropic to pull its most advanced models for nearly three weeks. Anthropic Chief Executive Dario Amodei called on the industry last month to slow development, warning that AI agents could control the whole internet within six to 12 months.

Charlie Bullock of the Institute for Law & AI called Anthropic’s proposal pretty reasonable and said it goes further than other frontier labs have done. He said on X that the FRONTIER Act, a competing House bill from this summer, is stronger than Anthropic’s proposal. An Anthropic spokesperson said the bill would grant the government the “ability to block or deter dangerous deployments of the most powerful models.”