Comment on the Bureau of Industry and Security’s Framework for Artificial Intelligence Diffusion
This comment was submitted by the Institute for AI Policy and Strategy (IAPS) in response to the Framework for Artificial Intelligence Diffusion (BIS-2025-0001), introduced by the Bureau of Industry and Security on January 15th, 2025. This comment references RIN 0694-AJ90.
Artificial intelligence (AI) is advancing at an extraordinary pace, promising to rapidly become integral to national security as AI is diffused across economies, militaries, and research institutions worldwide. In the coming years, countries with access to the most advanced AI models—and the capacity to deploy them broadly—stand to gain a significant strategic edge.
Training and deploying the most advanced models requires a lot of computational resources ("compute"). As a result, the most promising way to steer the diffusion of AI power towards America and its allies and partners and away from its adversaries is through the control of compute in the form of AI hardware and related infrastructure. Compute is currently a key bottleneck for every major AI company in every country, crucial not only for developing new generations of AI systems but also for deploying them widely and integrating them into the economy.
The Framework for Artificial Intelligence Diffusion (henceforth, the “diffusion rule”), introduced by the Bureau of Industry and Security (BIS), was an attempt to steer AI diffusion through the use of export controls. US export controls on AI chips are already having a tangible impact on China’s AI industry, as leaders of prominent Chinese AI companies, including DeepSeek, Tencent, and Alibaba, have acknowledged. Due to the restrictions, Chinese companies operate at a compute disadvantage to the US. US-led export controls have reduced China’s access to compute and increased its costs for the compute it can obtain. Analyses by Epoch AI found that, as of January 2025, only three out of 29 identified AI models trained at the scale of OpenAI’s GPT-4 were Chinese, and that the rate of scaling since the release of ChatGPT has been slower for Chinese models than American models. Not a single leading frontier AI supercomputer is Chinese.
However, while controls have been effective, they are still not as effective as they could be due to persistent enforcement failures. Smugglers have funneled substantial quantities of restricted AI chips into China. Additionally, despite sanctions, Taiwan Semiconductor Manufacturing Company (TSMC) has reportedly manufactured enough chip dies for Huawei to build a staggering one million Ascend 910C accelerators. When addressing this challenge, BIS should not only ensure that any new policies are enforceable but also that policies and technologies are put in place to address these enforcement gaps. And crucially, technologies like location verification can offer privacy-protecting, innovation-preserving ways to simultaneously streamline and enhance enforcement.
Any framework must carefully navigate the balance between robust security controls and the imperative to foster American innovation and maintain strong partnerships with allies. While the primary focus of this comment is on strengthening measures to protect US interests, we are mindful that overly broad or complex regulations can inadvertently stifle legitimate commerce or strain international cooperation. As the administration works towards a strong, streamlined successor to the diffusion rule, we offer recommendations for BIS across three core objectives:
Steer the global distribution of American compute to preserve America’s lead in AI
Ensure importing countries—including allies—uphold US export controls or face strict import limits, and use existing technology to address enforcement challenges such as illegal AI chip reexports
Secure key AI models stored on foreign soil, as model weight theft represents a major potential “compute shortcut” for adversaries