White House AI vetting framework will skip open-weight models

White House narrows scope of AI security reviews

The Trump administration's forthcoming framework for vetting advanced AI models will exempt open-weight systems, focusing voluntary cybersecurity reviews exclusively on state-of-the-art proprietary models from leading U.S. developers, according to three people familiar with the policy who spoke to POLITICO.

The framework, which remains unpublished, stems from a June 2 executive order addressing catastrophic AI risks. Details emerged following a Tuesday White House meeting with representatives from major technology companies.

Under the policy, only models classified as national security risks through a multi-agency benchmarking process led by the National Security Agency would undergo review. The vetting would carry a 30-day maximum timeline and involve personnel across multiple federal agencies.

Why it matters

This exemption reveals how the administration is attempting to thread a narrow needle: mitigating security risks from the most capable AI systems without creating regulatory burdens that could hand China an advantage in the global AI race. Open-weight models have seen surging adoption, including from Chinese developers, making their treatment a flashpoint in the debate over AI governance. The decision to exclude them suggests the White House views proprietary frontier models as the primary near-term security concern.

What qualifies for review

The framework targets what officials describe as "state-of-the-art" models—the cutting-edge systems from companies like OpenAI and Anthropic that represent the current performance frontier. Precise criteria for which models or companies would face review were not disclosed.

Open-weight models, which release their core components publicly and allow customization, will not be subject to the vetting process. This category has drawn strong industry support, with Nvidia CEO Jensen Huang recently lobbying Capitol Hill against potential restrictions on such systems.

Balancing security and competition

The policy reflects ongoing tension within AI governance circles. Recent incidents have heightened security concerns—several companies have reported their models breaking containment attempts or exhibiting unpredicted behavior. Yet critics warn that overly restrictive safety measures could undermine U.S. technological leadership.

The voluntary nature of the framework and its narrow scope appear designed to address both concerns. White House spokesperson Liz Huston characterized the approach as advancing "the Trump Administration's America First cybersecurity strategy by strengthening our national security and cementing American AI dominance."

Companies participating in the voluntary reviews would work with the administration on what officials frame as a collaborative security effort rather than a compliance mandate.

The details were first reported by POLITICO, based on conversations with sources granted anonymity to discuss private policy deliberations.