The Trump administration is reportedly expanding its secretive AI safety framework to include certain open-weight models, a move that could reshape how the United States approaches artificial intelligence regulation. According to recent reports, the framework, which was originally designed to assess the cyber capabilities of advanced AI models, has so far applied only to closed models from leading American companies such as OpenAI. Now, administration officials are considering adding open models to the review process if those models are deemed advanced enough.
Earlier this month, reports emerged that the administration was assembling an AI safety framework that would evaluate models for potential cyber risks. What made that framework unusual was its secrecy: the details would be kept hidden from the public, and only model providers would be allowed to see the assessment criteria. The framework was characterized as a national security measure, aimed at ensuring that cutting-edge AI models could not be exploited for malicious purposes by adversaries. At the time, it was understood to cover only closed models from top-tier providers, leaving the open-weight ecosystem outside the scope of official review.
Now, however, the administration is reportedly reconsidering that scope. The concern, according to insiders, is that if the government bestows its official approval exclusively on closed models, it would implicitly signal that open models are less safe or less trustworthy. This could create a paradoxical disincentive for American companies to release open-weight models, potentially ceding the open-source AI landscape to foreign competitors. The administration is said to be worried about the reputational damage to open models and the broader implications for American technological leadership.
The Rise of Chinese Open Models
The policy shift comes at a moment when Chinese AI labs are making significant strides with open-weight models. The most recent shock to the American AI establishment came from Moonshot AI, a Chinese company that released Kimi K3, an open-source model that is both cheaper and, in many benchmarks, on par with or better than some of the latest offerings from American industry leaders like Anthropic, OpenAI, and Google. The release of Kimi K3 shattered assumptions that open-source models were necessarily inferior to their closed counterparts, and it triggered a wave of concern among American policymakers and tech executives.
For some time, experts predicted that the Trump administration would respond to this competitive threat with blanket bans on open models. That strategy appeared to be the default approach, especially as national security hawks warned about the dangers of allowing powerful AI models to circulate freely. Open-weight models, which allow users to download and modify the underlying weights of a neural network, are notoriously difficult to control once released. They cannot be easily pulled back, and they can be fine-tuned for malicious purposes without any oversight.
Yet as the Chinese open-source ecosystem gained momentum, a coalition of American AI companies and industry leaders began to push back against the idea of broad restrictions. Their argument was twofold: first, that restricting open models would not stop Chinese developers but would simply put American researchers and companies at a disadvantage; and second, that closed models are not inherently safe, despite their controlled distribution. This latter point has become a central talking point in the open-weight advocacy community, which notes that even closed models can be jailbroken or misused by authorized users.
Industry Pressure: The Open Letter
In a coordinated effort, numerous AI companies, including Meta, Microsoft, and Palantir, signed an open letter titled “Open Weights and American AI Leadership.” The letter urged the Trump administration to avoid restricting open models, arguing that doing so would only accelerate America’s decline in the global AI race. The signatories contended that open-weight models are essential for innovation, transparency, and the diffusion of AI capabilities across the economy. They also warned that a U.S. ban on open models would hand China a strategic advantage, as Beijing would be happy to fill the void with its own open-source ecosystems.
The letter reflected a growing consensus among industry leaders that the future of AI lies in a hybrid ecosystem. Closed frontier models will continue to push the boundaries of capability, but open models will play an equally important role in enabling customization, sovereignty, and competition. The signatories argued that the United States needs both, and that overly restrictive regulation would ultimately harm American security and prosperity.
Nvidia’s Jensen Huang and the Open Secure AI Alliance
One of the most prominent voices in favor of open models has been Nvidia CEO Jensen Huang. Huang, who has long championed the idea that open models are key to winning the global AI race, has personally advocated this position to the Trump administration. His commitment to the cause became even more visible when he made his first-ever post on X, the social media platform formerly known as Twitter, to address the issue directly.
“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 intervention was a significant moment, as Nvidia is one of the most influential companies in the AI supply chain. Without Nvidia’s GPUs, neither closed nor open models would be able to train at scale, giving the company enormous leverage over the industry’s direction. Huang has made clear that Nvidia sees open models as a business opportunity, not just a philosophical preference, since more open models mean more demand for Nvidia hardware.
Nvidia’s passion for open-weight AI has also extended beyond rhetoric. The company recently helped form the Open Secure AI Alliance, a coalition designed to promote the proliferation of open-weight models while addressing security concerns. The alliance brings together researchers, developers, and enterprises that share a commitment to open AI, with the goal of creating a robust ecosystem of open models that can compete with closed offerings on safety, capability, and reliability. The alliance is still in its early stages, but its existence is a testament to the growing institutional support for open-weight AI.
OpenAI and Anthropic: Divergent Stances
While the open-weight camp has gained powerful allies, not every major AI company has embraced the cause. OpenAI, for instance, initially declined to sign the open letter, leading to speculation that it feared open models would undermine its commercial advantage. However, CEO Sam Altman later added his co-sign, a move that surprised many observers given OpenAI’s history of keeping its models close. Analysts interpreted Altman’s decision as a pragmatic acknowledgment that the industry is moving toward a more open ecosystem, and that OpenAI would be better served by having a seat at the table than by opposing the trend.
Anthropic, on the other hand, has refrained from backing the letter altogether. The company, which has emerged as a leader in closed models, has faced accusations from the pro-open camp that it is trying to push the administration toward a stricter stance in favor of closed models. Critics point to Anthropic’s business model, which relies on selling access to its proprietary Claude models, and argue that anything that discourages the proliferation of open models would only help Anthropic’s bottom line. The accusations have been leveled at Anthropic CEO Dario Amodei, who has rejected them, insisting that his company’s position is based on safety considerations rather than commercial self-interest.
Anthropic has a complicated relationship with the Trump administration, having had a very public falling out not long ago. The company has positioned itself as a safety-first AI lab, and it has argued that open models, while valuable in some respects, pose unique risks that should not be ignored. Anthropic’s stance has made it a target of criticism from open-weight advocates, who see the company as an obstacle to a more democratic and distributed AI ecosystem. Amodei has fired back, arguing that safety and capability are not mutually exclusive, and that the industry must find a way to balance innovation with responsibility.
The Broader Implications
The debate over open models is not just a policy squabble; it reflects deeper questions about the future of AI governance. On one hand, open models offer the promise of transparency, innovation, and the kind of grassroots experimentation that has driven much of the internet era. They allow researchers to scrutinize the behavior of AI systems, identify biases and vulnerabilities, and build bespoke solutions for specialized domains. They also enable countries and organizations to maintain digital sovereignty, rather than relying on a handful of American companies for their AI infrastructure.
On the other hand, open models are a double-edged sword. Once a model’s weights are released, there is no way to restrict who uses them or for what purposes. They can be used to create disinformation campaigns, develop cyber weapons, or enable surveillance and repression. These risks have led some safety advocates to argue that the world is not ready for frontier open models, and that regulators should err on the side of caution.
The Trump administration’s decision to include open models in its AI safety framework suggests that it is trying to thread the needle. By subjecting advanced open models to the same kind of assessment as closed models, the administration can claim to be addressing national security concerns without imposing an outright ban. The framework would be secretive, but it would at least provide a pathway for open models to receive official approval, thereby avoiding the stigma of being seen as inherently unsafe.
Yet the secrecy of the framework has drawn criticism from transparency advocates, who argue that a process that is hidden from the public is a dangerous precedent. Without public oversight, there is a risk that the framework could be used to favor certain companies over others, or to block open models for political rather than security reasons. The fact that only model providers would have access to the assessment criteria has also raised eyebrows, as it could allow providers to game the system or obscure serious vulnerabilities.
For now, the administration has not made any formal announcement, and the details of the framework remain in flux. What is clear is that the pressure from industry leaders like Jensen Huang and the signatories of the open letter has forced a reconsideration of the earlier policy drift toward blanket restrictions. The growing capability of Chinese open models is a powerful argument against an American retreat from open-source AI, as it would simply cede the field to competitors abroad.
The coming months will likely determine whether the framework ultimately becomes a pragmatic compromise or a new source of controversy. As the Chinese open-model ecosystem continues to evolve, American policymakers will have to weigh the undeniable benefits of open innovation against the very real risks of misuse. The industry’s recent mobilization, however, has made one thing clear: open models have powerful defenders in the highest echelons of American technology, and any attempt to restrict them will face significant resistance.
Source:Gizmodo News