Artificial intelligence is at the center of one of the most heated debates in American politics in 2026.
And while warnings keep piling up, the White House remains silent about a document that practically everyone wants to see.
The framework created by the U.S. government to test the most advanced AI models in the world was finalized back in August, but so far no one outside the government knows exactly what it says. No criteria have been released. No results have been published. And no date has been given for that to change.
What there is no shortage of, though, are the warnings. Current and former employees at Anthropic, one of the largest companies in the field, are raising their hands to say that AI could pose a real risk to humanity before this decade is over. In a recently released report, Anthropic itself revealed that it blocked multiple attempts by scientists to use its Claude model for research that could lead to the creation of bioweapons. There is a lot happening all at once, and the question hanging in the air is simple: how much longer can the government keep all of this out of the public eye? 🤔
What is this framework and why does it matter so much
When we talk about an artificial intelligence evaluation framework, we are talking about a structured set of criteria, tests, and metrics designed to determine whether an AI model is safe enough to be released. Think of it as a kind of technical assessment report, except applied to systems that can make decisions, generate content, and depending on their level of access, influence critical areas like healthcare, national defense, and infrastructure.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
This framework is voluntary, and it came out of an executive order signed by President Trump in June. The core idea was to ask companies like Anthropic and OpenAI to give the federal government access to their most advanced models up to 30 days before launch, with the goal of strengthening the safety of these systems. Worth noting a curious detail here: Trump actually scrapped an earlier version of this executive order, claiming he did not want the federal government slowing down innovation. So from the very beginning, there was already tension between safety and the speed of development.
The central problem here is not just the lack of access. It is what that lack of access represents within a context where the very companies building these systems are publicly admitting the risks are real and immediate. Since the document remains confidential, no one knows for sure what standards the government is asking companies to meet, or whether these companies are actually disclosing their latest breakthroughs. And there is an added wrinkle: neither the government nor the companies are required to publish the results of these evaluations, or even confirm whether they participated in the process.
The absence of transparency creates an information vacuum that, on a topic this sensitive, is anything but reassuring. If the government has a document that defines how these risks should be assessed, and that document is completely locked away, it becomes impossible to know whether the criteria being used are rigorous enough or whether the models already in operation have gone through any meaningful evaluation.
Safety in question: the warnings that keep coming
Anthropic is one of the most respected companies in the artificial intelligence space, and the fact that it publicly disclosed blocking attempts to use Claude for research related to bioweapons was not a marketing move. It was a direct warning about where the boundaries are being tested right now, not in some distant future. When a company that literally builds one of the most advanced models in the world says it had to step in to prevent a potentially catastrophic use case, the bare minimum anyone should expect from a responsible government is that the criteria used to evaluate these systems be known and auditable.
And the warnings are coming from inside the house too. Recently, an Anthropic researcher and former OpenAI employee named Jacob Coxon resigned and posted something on social media that should give anyone pause: the people building AI genuinely believe it could end all of us by the end of the decade. Evan Hubinger, another Anthropic employee, publicly agreed, reinforcing that they truly believe AI could kill every human being. These are not lines from a science fiction movie. These are statements from people working at the heart of this technology. 😳
On the other side, OpenAI seems to be taking the collaborative route. CEO Sam Altman said the company submitted its powerful new model, Astra, for review and described the process as productive. According to him, as these models reach significant levels of capability, close engagement with safety institutes in the United States, the United Kingdom, and elsewhere around the world becomes increasingly important. It is a positive signal, but it also blows wide open the same question: if the process is that important, why are the criteria still hidden?
The safety question in artificial intelligence is not limited to doomsday scenarios. It shows up in very concrete decisions: which models can be used in medical applications, which ones get access to critical infrastructure, which ones can be integrated into defense systems. If the criteria defining those permissions live in a closed document, with no external review and no timeline for change, the risk is not just technical. It is institutional. And that kind of risk tends to be much harder to fix once it actually becomes a problem.
Transparency: the debate the government cannot avoid
The discussion around transparency in artificial intelligence policy is not new, but it has never been as urgent as it is right now. In other regulated industries, like pharmaceuticals and finance, there is a clear logic: whoever approves and whoever gets approved must follow criteria that can be audited, questioned, and when necessary, challenged. With AI, the logic should be the same, especially when we are talking about systems with capabilities that go far beyond a simple recommendation algorithm.
Civil society organizations are pushing for action. The Center for Democracy and Technology sent a letter to the White House asking the Trump administration to release its framework for reviewing frontier models. Tim Harper, who works at the organization focusing on election security, summed up the issue well when he said the public has the right to know what the government’s interpretation of the law is on these topics. He also pointed out that this reflects a broader lack of transparency from the administration in revealing internal discussions and conversations.
And this movement is not isolated. In July, more than a thousand employees from AI companies signed a public statement calling on the U.S. government to support an international effort to develop the technical and governance tools needed to deliberately slow down the pace of automated AI development. When the people building the technology are asking for the brakes, maybe it is time to listen carefully.
On the government side, though, the priority seems to be elsewhere. The White House has been pressing forward with support for advanced AI model development, even in the face of internal and external warnings, driven by the belief that the United States needs to achieve global dominance in artificial intelligence and outpace China. Trump has been a strong advocate for AI development and building data centers domestically, going so far as to say that China could not be happier with any opposition to new American infrastructure.
Meanwhile, the public appears divided and cautious. Recent CBS News polls found that most Americans believe AI will eliminate jobs in the country, and nearly two-thirds think government policy probably will not ensure AI is used appropriately. That is a clear message: trust in the government’s ability to regulate the technology is far from solid.
The picture taking shape is concerning not because the government is evaluating AI models, but because it is doing so in a completely opaque way, at a moment when the pressure for accountability has never been higher. Companies in the sector are growing in speed and influence. The models are becoming more capable with every development cycle. And the document that should guide how all of this is evaluated remains locked away, with no timeline for release, while the world outside debates the very risks it was supposedly created to address.
Reached repeatedly for comment on the matter, the White House did not respond. And it is precisely that silence that keeps the discussion alive. Transparency, in this context, is not just a political demand. It is a technical and ethical necessity that the government’s silence makes harder and harder to ignore. 🔍
