Trump Administration Expands AI Safety Framework to Include Open AI Models Amid China Competition

Trump Administration Expands AI Safety Framework to Open AI Models

The Trump administration is expanding its secretive AI framework to include advanced open AI models that reach frontier-level capabilities, marking a significant shift in how the government approaches AI oversight1

1

. The voluntary review process, which initially applied only to closed models from providers like OpenAI and Anthropic, will now encompass open-weight AI models deemed sufficiently advanced2

2

. This White House AI testing shift addresses concerns that excluding open models from government scrutiny could paradoxically harm their reputation and disincentivize American companies from releasing them1

1

.

The AI safety framework defines covered frontier models as those with state-of-the-art capabilities and national security risks2

2

. Under the June executive order, developers can provide the federal government with early access for cybersecurity evaluation through pre-release reviews before broader release3

3

. Some companies consulted during discussions actually requested to be subject to this voluntary review process2

2

.

Industry Pushback Shapes Policy Direction

The policy evolution follows intense lobbying from major AI companies who view open models as critical for U.S. competitiveness. Meta, Microsoft, and Palantir signed an open letter titled "Open Weights and American AI Leadership," urging the Trump administration to avoid restricting open models1

1

. The letter argued that limitations would cause the U.S. to fall behind as China races ahead, and claimed closed models are "not inherently safe." Nvidia CEO Jensen Huang made his first-ever X post championing this cause, stating that "open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty"1

1

. Nvidia subsequently formed the Open Secure AI Alliance to support open-weight AI models proliferation.

OpenAI CEO Sam Altman later endorsed the letter, though Anthropic refrained from backing it1

1

. Critics accused Anthropic CEO Dario Amodei of pushing for stricter stances favoring closed models to benefit his company's business, though Amodei rejected these accusations. Treasury Secretary Scott Bessent and White House Office of Science and Technology Policy Director have recently posted support for open-source AI faces more government scrutiny2

2

.

Chinese Open Models Trigger Strategic Reassessment

The administration's approach shifted after Chinese open models demonstrated they were catching up to American closed offerings. Last month, Moonshot released Kimi K3, an open-source model cheaper yet comparable to or better than latest releases from Anthropic, OpenAI, and Google1

1

. This development sparked concerns about U.S. competitiveness in the geopolitical context of AI development. The Trump administration has consistently viewed anything at or above the capabilities of Anthropic and OpenAI's most advanced models as needing government collaboration, regardless of whether they're open- or closed-source2

2

.

Implementation Timeline and Risk Management Implications

A national security presidential memorandum issued in June could separately set standards for government use of open models2

2

. By early September, officials must issue new policy governing AI use in national security systems. By early October, officials are tasked with developing standardized methods for AI evaluation and baseline security and risk management practices2

2

. According to Eric Syphard, Booz Allen's head of AI, these October deadlines could clarify how open-weight models will be evaluated and procured for national security use2

2

.

For enterprises, particularly financial institutions, the distinction matters significantly. Open-weight models can be downloaded, customized and hosted privately, making them appealing to banks and retailers wanting control over cost, data location and model behavior3

3

. However, once powerful open models are released, safeguards can be modified or removed by downstream users, creating different risk profiles. The regulatory uncertainty means institutions will need more rigorous compliance processes as they weigh whether federal testing can adequately address enterprise-specific security needs when deploying models to democratize AI access while maintaining operational controls.