Some U.S. tech industry figures and members of Congress argue that the mechanism discussed between the White House and AI companies has clear shortcomings in transparency, long-term stability, scope, and risk-handling procedures, with many important questions still unanswered. For example: at what capability level does a model fall into the evaluation scope; what specific metrics and thresholds are used; are uniform standards applied across companies; would a company face consequences for refusing to participate; and what measures can be taken if a model is found to carry severe risks?

Analysts say the mechanism does not currently constitute a mandatory approval system, nor does it carry public enforcement tools. But for companies hoping to participate in government projects, this voluntary evaluation may gradually become an industry norm.

For developers of closed-source models such as OpenAI, Google, and Anthropic, if their models cross the relevant capability threshold and they choose to participate, they may need to engage with the government earlier—raising costs in model confidentiality, access-rights management, and release-plan coordination. If comparable open-weight models do not undergo evaluation, this could invite criticism over inconsistent regulatory standards. It also remains unclear which institutions can be recognized as “trusted partners,” and whether foreign governments or overseas entities might be included.