America is in an AI race that will shape our economy, our national security, and our future. But you cannot win a race by giving some competitors the rules and leaving everyone else guessing.
The White House has finalized its new AI safety framework, but it does not plan to publicly release the framework or the evaluation criteria it will use for advanced AI models. Instead, only a handful of large AI companies will be informed of the standards and can therefore act accordingly. Despite clearly good intentions to protect national security while fostering growth, this decision is the worst of both worlds, pairing all the competition-slowing harm of nebulous governance with the trust-destroying lack of transparency and favoritism.
Under the new 30-day pre-release review process for frontier AI models, the evaluation benchmarks and thresholds reportedly will not be publicly disclosed. That means researchers, startups, open-weight developers, and many businesses may not know how these systems are being judged.
That isn’t governance. That’s exposure.
If this framework will influence how the most powerful AI systems are evaluated, the public, researchers, businesses, and our allies deserve to understand the principles behind it. By only giving the details to a select number of favored companies, the administration is effectively picking winners and losers in the AI space, giving some companies a cheat sheet while holding others to obscure rubrics.
National security and transparency do not need to be mutually exclusive.
In an effort to preserve national security while staying competitive in the AI race against China, this playing favorites hamstrings American innovation while not providing the fair and transparent governance and guardrails that many have been calling for consistently.
A year ago, we were told that governance and guardrails meant losing to China. Now, as China gains a definitive lead, guardrails are being implemented in a manner that seems tailor-made to slow down the American AI sector, stifle competition, and guarantee a loss.
The Trump administration has an opportunity to shape policy in a way that actually fosters growth, healthy competition, and a flourishing tech sector by applying the framework universally and allowing all American companies to have the same amount of knowledge, therefore playing on a level field. But if this decision to keep the rules private but give some companies inside access remains, transparency and competition are fundamentally threatened.
Without transparency, trust is lost. Innovation flourishes when expectations are clear, standards are transparent, and everyone understands the rules of the road. When the largest AI companies help shape a framework that isn’t open to broader scrutiny, there’s a legitimate concern that it could create the perception — or reality — of an insider’s club.
One of the oldest principles of security is verification. It’s difficult to build public trust when the safety tests themselves remain hidden.
The national security argument is compelling, and I understand the administration’s concern. Some evaluation methods may reveal capabilities that could benefit foreign adversaries or expose sensitive cybersecurity techniques.
Protecting the U.S.’s national security is essential, but secrecy cannot and must not become a substitute for good governance.
American leadership has long been built on open standards, independent verification, competition, and public trust. National security and fair, transparent standards are not a zero-sum game. We can have both.
Rather than emphasize one at the expense of the other, the United States needs a balanced approach to ensure that frontier AI models are not risks without sacrificing the governance and guardrails that keep the country safe or the competition that keeps us ahead of China.
We can protect genuinely sensitive testing methodologies while maintaining transparency by publishing high-level evaluation standards and governance principles and keeping the detailed specifics vague. Expectations must be clear and easily provided, so that startups, enterprises, and open-weight AI developers understand the rules. Rather than the federal government overseeing the testing, which could lead to huge backlogs and inefficiencies, we ought to allow independent, third-party auditors to validate compliance.
If artificial intelligence is going to reshape our economy, our national security, and our daily lives, public trust won’t come from closed-door decisions alone. It will come from secure and transparent governance.
THE SOVIET UNION SPENT BILLIONS INFILTRATING THE WEST. RUSSIA JUST BOUGHT A $20 AI SUBSCRIPTION
We were told that guardrails would make America lose to China. Now we risk losing because of the way we are building them.
We don’t need fewer guardrails. We need better ones.
Theresa Payton served as the first female White House chief information officer under President George W. Bush. She is the CEO of the cybersecurity firm Fortalice Solutions.