White House Weighs Expanding AI Vetting to Open-Weight Models

The Trump administration is considering a significant pivot in its artificial intelligence strategy by expanding government vetting to include open-weight and open-source models. This potential shift marks a departure from early policy goals aimed at maximizing U.S. innovation through deregulation.

In early August 2026, the White House introduced an AI safety review framework that explicitly exempted open-weight models developed by U.S. companies from government oversight. However, officials are now grappling with how to regulate these models to mitigate security risks. This reconsideration comes as the administration balances the need for domestic agility against the rise of powerful open-source AI from global competitors, specifically China.

The Security Dilemma

The shift in approach is driven by the inherent nature of open-weight models, which allow users to download and modify the underlying calculations of the AI. Unlike closed-system APIs, where a company maintains total control over the model's guardrails, open-weight models can be stripped of safety filters once released. This creates a persistent security vulnerability that the administration is now seeking to address.

Adding to the tension is the government's own struggle with transparency. The White House has declined to publicly release the AI model evaluation framework it is using, a decision following a series of security breaches at leading AI labs OpenAI and Anthropic.

Industry Implications

If the administration moves forward with this expansion, the impact on the AI ecosystem will be profound. Companies that have championed the open-source movement will face new regulatory hurdles. Meta, for instance, released Muse Glimmer, an open-weight agentic model, in August 2026; it exemplifies the type of high-capability software that would fall under a broader vetting regime.

Historically, the U.S. government has struggled to reconcile the promotion of AI leadership with national security. While the previous administration's Executive Order 14110 tasked the NTIA with reviewing open-weight models, the current administration initially leaned toward a more permissive environment to maintain a competitive edge over foreign adversaries. The current pivot suggests that the perceived risks of uncontrolled open-source AI may now outweigh the benefits of total deregulation.

What to Watch

Industry observers are now waiting to see if the White House will implement a tiered system of review or a blanket requirement for all high-capability models, regardless of their distribution method. It remains unclear how the administration will define the threshold for "security risks" that trigger government review, or how it will handle models that are developed domestically but modified by foreign actors after release.