As AI labs continue to make headway towards highly advanced AI, governments have to keep pace with chip export controls beyond hardware restrictions in order to prevent their misuse.
The U.S. Bureau of Industry and Security (BIS) controls advanced chips through performance thresholds, blocks military uses through end-use restrictions, and applies Foreign Direct Product Rule to manufacturing equipment. After the AI Diffusion Rule was rescinded in May 2025, BIS's chip export controls prioritise easier chip access for American allies and greater restrictions for China. These controls are not designed to track chips after export. China has reportedly acquired tens of thousands of H100-class chips through indirect channels (CNBC). Chips designed for training can be repurposed for inference, and evolving AI architectures change which chip specifications are most valuable. There is usually a 6-12 month lag between when a gap in export control is identified and a new policy is implemented and becomes effective.
BIS should develop tiered export licensing and create a technical assessment team at CAISI to strengthen chip export controls. A tiered approach would restructure controls based on end users' willingness to accept verification. For example, Tier 1 would provide expedited licensing for advanced chips (e.g: B200) to end users who accept quarterly aggregate compute usage reporting, remote chip inventory verification, and network isolation. At Tier 2, chip end users would be subject to on-site inspections. At the 3rd Tier, end users who refuse verification would be denied chips. To monitor this, BIS should establish a small technical team (5-7 staff) in NIST's AI Safety Institute to stress-test regulations against emerging AI developments. This 'regulatory red team' could reduce the current 6-12 month implementation lag to just several weeks by continuously stress-testing controls against new AI developments.