WASHINGTON — The White House is refusing to publish the voluntary framework governing the review of frontier artificial intelligence models, a decision that has drawn sharp criticism from smaller AI labs and open-source developers who fear the secretive process will entrench dominant corporate players at the expense of American innovation and national sovereignty.
The framework, developed over two months following a June executive order, was discussed Tuesday at a White House meeting with a select group of companies including OpenAI, Anthropic, Google, Meta, and Nvidia. The terms of the agreement remain hidden from the public and much of the industry, including critical details such as the list of "trusted partners" granted early access to the world's most powerful models and the specific thresholds that trigger a government review. An administration source confirmed only a handful of firms were in the briefing room.
Classified Benchmarks and Closed Doors
The executive order itself laid the groundwork for opacity, stipulating that the benchmarking process to designate a "covered frontier model" may be classified, with final determination resting with the Director of the National Security Agency. Reports indicate the covered models are defined as closed-source, possessing state-of-the-art capabilities and presenting national security risks—terms a source familiar with the briefing said remain dangerously undefined. Companies can voluntarily submit such a model for a 30-day government review before release, during which access is heavily restricted in high-security environments.
For American economic sovereignty, the framework's scope is a critical question. Open-weight models have reportedly been excluded, a carve-out that some argue will benefit smaller U.S. firms chasing market leaders by reducing their regulatory burden. However, this exclusion simultaneously leaves outside the framework the very models that pose the most direct competition to American primacy: the open-weight releases from Chinese state-aligned entities like Alibaba, DeepSeek, and Moonshot AI that consistently approach the American frontier.
The unspoken concern is a government-backed moat around a few favored corporations, pushing the market toward 'government-approved' closed-source models and away from the open ecosystem that fuels American dynamism.
Market Distortion and Worker Impact
The framework's structure introduces potential for significant market distortion. By creating a class of government-vetted models with early access for unnamed "trusted partners," Washington risks channeling capital and customer adoption toward a small cartel of Silicon Valley giants, locking out domestic start-ups. This consolidation directly impacts American workers. An AI market dominated by a few closed-source providers stifles the diffusion of productivity-enhancing tools across the broader domestic economy, concentrating high-wage technical jobs within a narrow corporate band rather than distributing them across the industrial base.
Furthermore, the reported restriction on a company's own employees accessing their submitted model during the 30-day review window raises national security and commercial viability questions, potentially delaying American firms' ability to deploy and iterate on cutting-edge technology while foreign adversaries face no such domestic hindrance.