The Trump administration is expanding its secretive AI safety framework to inspect some open models before they are made public, according to multiple reports. The framework, which is intended to assess the cyber capabilities of advanced AI systems, has so far been applied only to closed models from top providers such as OpenAI. But open models deemed advanced enough are now expected to fall under the same review process.

Key facts

• The Trump administration's AI safety framework will likely expand to cover advanced open models.
• The framework assesses cyber capabilities and is kept secret from the public.
• Chinese open models such as Moonshot's Kimi K3 are driving renewed U.S. interest in open weights.
• Major U.S. tech firms signed an open letter urging the administration to support open models.
• Nvidia CEO Jensen Huang publicly backed open models and helped launch the Open Secure AI Alliance.
• Anthropic declined to sign the letter, prompting accusations against CEO Dario Amodei.

A secretive framework with wide-reaching implications

Earlier this month, reports emerged that the White House was assembling an AI safety framework designed to evaluate the cyber capabilities of cutting-edge models. According to those reports, the entire process would be kept secret, with details shared only with model providers. The administration argued that secrecy was necessary to protect national security, but critics say it creates a dangerous lack of accountability around some of the most influential technology in the world.

The framework initially targeted closed models from leading American AI companies, including OpenAI and others with large deployments. Closed models are those whose underlying weights and architecture are kept private, with users accessing the technology through an API or subscription. The new reports indicate that the framework will soon include open models, which are released with their weights available for developers to download, modify, and deploy on their own infrastructure.

The policy shift appears to be driven by a concern within the administration that approving only closed models would harm the reputation of open models. If Washington's official safety seal was reserved for closed models, the thinking goes, open models would be perceived as riskier or less trustworthy. That would create a 'paradoxical disincentive' for American companies to release their own open models, according to people familiar with the matter.

Why open models have become a national security question

Open-weight models have long been a divisive topic in Washington. Proponents argue that open models encourage transparency, allow outside researchers to audit the technology, and make advanced AI available to smaller businesses, universities, and governments around the world. They also point out that closed models are not inherently safe, and that secrecy can conceal vulnerabilities rather than eliminate them.

Skeptics warn that open weights can be copied, fine-tuned, and weaponized by malicious actors who do not have access to the same resources as frontier labs. A bad actor can download an open model and remove safety alignments, then use it for disinformation, cyberattacks, or other harmful purposes. Because the weights are public, it is almost impossible to claw them back once released.

The debate has become more urgent as Chinese AI labs have begun releasing open models that match or exceed the performance of American closed systems. Last month, Moonshot released Kimi K3, an open-source model that was significantly cheaper than comparable offerings from Anthropic, OpenAI, and Google, while matching or beating them on several benchmarks. The release triggered a wave of concern in Washington, where some policymakers saw it as evidence that the United States was losing its edge in AI.

Industry leaders push back against restrictions

For a time, the Trump administration appeared to be leaning toward broad restrictions on open models, including potentially requiring government approval before any advanced open-weight model could be released. That approach alarmed AI executives, who argued that such a policy would hand China an advantage in the global race for AI dominance.

Several major companies, including Meta, Microsoft, and Palantir, signed an open letter titled 'Open Weights and American AI Leadership.' The letter urged the administration to avoid restricting open models and argued that doing so would only cause the U.S. to fall behind as China races ahead. The signatories also pushed back against the assumption that closed models are safer, noting that closed systems have their own security and accountability problems.

OpenAI, which initially did not sign the letter, later added its co-signature through CEO Sam Altman. The decision reflected a broader recognition within the industry that open models are an important part of the United States' AI ecosystem, even for companies that primarily sell proprietary systems. It also signaled that the debate over open versus closed AI is not a simple fight between industry factions, but a complex policy question with major strategic consequences.

Nvidia and Jensen Huang join the cause

Nvidia CEO Jensen Huang has been one of the most visible voices in favor of open models. Huang has long argued that open models will be key to winning the global AI race, and he has reportedly made that case directly to the Trump administration. His advocacy went public in a notable way when he made his first-ever post on X, the social media platform formerly known as Twitter.

'Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty,' Huang wrote. 'The world needs both frontier closed models and frontier open models.'

Huang's involvement extended beyond social media. Nvidia helped form the Open Secure AI Alliance, a coalition of companies and research organizations created to support the proliferation of open-weight AI models. The alliance aims to provide security tools, best practices, and shared infrastructure to make open models safer and more widely usable. Nvidia's role is significant because the company makes the GPUs that train and run most advanced AI models, giving it enormous influence over the industry.

Anthropic refuses to back the letter

Not every major AI company joined the push to support open models. Anthropic, one of the leading developers of closed frontier models, refrained from backing the open letter. The decision immediately drew accusations from the pro-open model camp, with critics claiming that CEO Dario Amodei was trying to push the administration toward a stricter stance that would benefit Anthropic's business.

The accusation stems from a straightforward business logic: if regulations make it harder to release open models, companies that sell closed models stand to gain market share. Anthropic, which has positioned itself as a safety-focused AI company, would be a clear beneficiary of such a dynamic. The company had a very public falling out with the Trump administration not long before the open letter was circulated, which complicated its position even further.

Amodei has rejected the accusation, insisting that Anthropic's position is based on genuine safety concerns rather than commercial self-interest. Industry analysts remain divided over whether Anthropic's reluctance reflects a principled stance on AI safety or a competitive calculation. What is clear is that the open letter exposed a growing rift in the American AI industry, with companies like Meta and Nvidia on one side and Anthropic on the other.

What expanding the framework could mean

Expanding the secretive AI safety framework to include open models would mark a significant change in U.S. policy. For open-model developers, the new process could provide an official approval pathway that validates their work and makes it easier to deploy their models in government and enterprise settings. It could also create new compliance costs and delays, however, particularly for smaller labs that lack the legal and technical resources of major companies.

For the administration, the move is something of a balancing act. It wants to prevent Chinese AI models from surpassing American systems, but it also wants to maintain the openness that has driven U.S. innovation for decades. The inclusion of open models in the safety framework is an attempt to have it both ways, offering a path to legitimacy for open models while still maintaining a degree of government oversight.

The details of the framework remain murky. It is not yet clear how the administration will determine which open models are advanced enough to require inspection, or what standards the process will use. Even the basic mechanics are unknown: how long the review will take, who will conduct it, and what type of findings will be shared with the public. The lack of transparency has drawn criticism from civil liberties groups, researchers, and even some industry executives who support the idea of safety reviews but object to doing them behind closed doors.

What is increasingly certain is that the Trump administration is moving toward a more structured approach to AI governance. The decision to include open models in the secretive framework suggests that officials have listened to industry arguments about the importance of open weights, while still seeking a mechanism to address national security concerns. The outcome of this debate will shape the future of American AI leadership, determining whether the country remains the world's center of AI innovation or cedes that position to China.

Source: Gizmodo News