Gottheimer proposes mandatory AI model reviews for security risks

Rep. Josh Gottheimer (D-N.J.) is preparing legislation that would mandate government review of powerful artificial intelligence models before deployment, requiring companies to submit their systems for screening of national security, critical infrastructure, cybersecurity, and bioterrorism risks.

The proposal, details of which Gottheimer shared exclusively with POLITICO, represents one of the most aggressive regulatory approaches yet from a key congressional AI policymaker. It comes as Washington grapples with concerns over advanced models like Anthropic's Claude Mythos, which some fear could enable bad actors to engineer dangerous bioweapons or superviruses.

Why it matters

The mandatory review framework would mark a sharp departure from the voluntary approach favored by both the White House and a competing bipartisan House proposal. If enacted, it would give federal agencies direct oversight authority over AI development at a time when the technology is advancing faster than existing regulatory structures can accommodate. The proposal also signals growing Democratic skepticism that industry self-regulation can adequately address catastrophic risks.

Parallel efforts and political tensions

Gottheimer's bill would run parallel to a bipartisan House effort led by Reps. Jay Obernolte (R-Calif.) and Lori Trahan (D-Mass.), who unveiled a discussion draft in June. That framework would require top developers to disclose safety and security risks but relies on voluntary model evaluations through the Commerce Department's National Institute of Standards and Technology.

The Democratic commission Gottheimer co-chairs—convened by House Minority Leader Hakeem Jeffries to develop the party's official AI policy agenda—swiftly criticized the Obernolte-Trahan draft, saying it failed "to meet the enormity of the moment."

Gottheimer told POLITICO that threats identified from models like Anthropic's Mythos "highlighted how critically important it is that we have a mandatory process for the government to review advanced models." The Trump administration imposed export controls on Anthropic's latest models on Friday over national security concerns.

Legislative path forward

The New Jersey Democrat, a moderate who has positioned himself as eager to reach bipartisan agreement on AI regulation, said his proposal is currently under review by the House Legislative Counsel to ensure consistency with existing laws. He is speaking with both Democrats and Republicans to build support.

The competing approaches reflect deeper tensions over how aggressively to regulate AI development. While Gottheimer's mandatory review framework would give government direct gatekeeping authority, the voluntary approach in the Obernolte-Trahan bill would preserve more industry autonomy while potentially preempting some state AI laws.

Details were first reported by POLITICO.