Imagine boarding a plane and being told, “Don’t worry, the pilots have a checklist. It’s just classified.” You’d probably get off the aircraft. Yet that is exactly what the White House is asking the American public to do with the most consequential technology of our lifetime.
According to Axios, the White House does not plan to publicly release its new framework for evaluating advanced AI models. Three sources familiar with the discussions confirmed it. The official rationale? Security. The actual rationale? Something far more uncomfortable.
When a government refuses to show its work on a matter of public safety, it isn’t protecting secrets—it’s protecting itself.
Let’s be clear about what we’re discussing. This isn’t a black-ops military program or a counterterrorism protocol. This is the benchmark for evaluating whether advanced AI systems are safe to deploy. It determines whether a model can be released to the public, integrated into infrastructure, or sold to enterprises. It is, effectively, the safety inspection for the digital engines that will power our economy.
And you don’t get to see the criteria.
Proponents will argue that there are legitimate reasons for secrecy. They’ll point to national security concerns, the risk of adversarial nations gaming the system, and the need to protect proprietary information. These are not trivial concerns. But they are also not the whole story.
The uncomfortable truth is that if this framework were defensible, a declassified summary would be easy. The White House could release a redacted version, an overview of the risk categories, or even a simple statement of principles. They have done none of this. The silence is the answer.
Secrecy turns a technical evaluation into a political trust test—and the administration knows they would fail.
Here is what the administration is trying to avoid: accountability. If the framework is made public, every decision becomes reviewable. If an AI model is approved and later causes harm, the public can point to the criteria and ask why it passed. If a model is rejected, the developer can argue against the standards. By keeping the framework secret, the White House retains absolute discretion. They can approve models that align with their interests and reject those that don’t, all without explanation.
This is not about safety. It is about power.
Think about the implications for you. If you use AI tools, build with them, or simply worry about where this technology is heading, you are now operating in the dark. The people making decisions about AI risk have decided that you do not need to know how they define risk. They have decided that your input is irrelevant.
This sets a dangerous precedent. The market demands transparency because it is a fundamental mechanism of trust. When that mechanism is removed, trust evaporates. And in its place, we get speculation, fear, and cynicism. We get a public that assumes the worst because the alternative—that the criteria are simply indefensible—is too grim to accept.
If you cannot defend your standards, you don’t have standards—you have preferences.
There is a way forward. The administration could release a framework summary that outlines the risk categories without revealing sensitive thresholds. They could establish an independent oversight board with security clearances to audit decisions. They could commit to publishing anonymized case studies of how the framework has been applied. These are not radical proposals. They are the minimum bar for democratic accountability.
But don’t hold your breath. The current trajectory suggests that the White House would rather preserve its flexibility than build public trust. That is a choice. And like all choices, it comes with consequences.
The consequence is that when the first major AI incident occurs—and it will—the public will not ask what went wrong. They will ask what was hidden. And the administration will have no answer.
That is the price of opacity. And eventually, we all pay it.
FAQ
Q: Could there be legitimate national security reasons for keeping the AI framework secret?
A: Yes, but they don't justify full opacity. The government can protect sensitive thresholds while releasing a summary of risk categories, decision principles, and oversight mechanisms. If they can't share even that, the problem isn't security—it's political vulnerability.
Q: What is the practical impact of this secrecy on AI developers and users?
A: Developers can't know if their models will pass review, creating uncertainty that stalls innovation and funnels power to those with political connections. Users get no way to assess the tools they're adopting. Everyone operates in the dark, and the government gets to move the goalposts without explanation.
Q: Is full transparency actually feasible for AI evaluation?
A: Feasible and necessary. The government doesn't need to publish every technical detail. It needs to publish the principles, the categories of risk, the decision-making process, and an independent audit mechanism. That's how every other safety-critical industry works—from aviation to pharmaceuticals. AI shouldn't be the exception.