China's AI Export Controls Target High-Performance Models

China's draft rule restricts advanced AI exports to non-approved nations, setting technical thresholds for models and tools.

China’s Ministry of Commerce and Ministry of Science and Technology released a draft export control rule on October 16, 2026. The rule targets 23 specific AI technologies, including frontier models, AI agents, and autonomous malware tools. It imposes restrictions on AI models trained on Chinese 'important data' unless the data meets three strict conditions: it must be cleaned beforehand, never appear in AI outputs, and constitute ≤0.1% of model inputs without substantially boosting core R&D. For models themselves, restrictions apply if they reach 2 trillion parameters, require ≥10^26 FLOPs of training compute, or rank in the global top 10 based on public metrics from major evaluators. AI tools capable of independently writing or altering malware to attack critical infrastructure face a total export ban, while vulnerability detection tools face restrictions but not bans. Open-weight models published via open-source releases are exempt but require registration with the Ministry of Commerce for automatic licensing.

The rule creates a technical barrier for non-approved nations, particularly those lacking Chinese data compliance infrastructure or access to high-performance computing. China’s fast-track process for Chinese firms operating overseas data centers or foreign research partnerships offers limited pathways for its own ecosystem. While China pledges to support AI development in the Global South through its World Artificial Intelligence Cooperation Organization, this initiative operates within China’s defined framework and faces constraints from the rule’s technical thresholds.

This move restricts the global spread of high-performance AI capabilities, potentially slowing adoption in regions without Chinese technical infrastructure or data compliance capacity. For the Commons, it increases friction for nations seeking advanced AI without meeting China’s stringent data and compute thresholds. The impact on abundance depends on whether Global South partners can bypass these barriers through China’s cooperative mechanisms or develop independent capabilities.

The rule remains draft status until final implementation. Top-10 rankings rely on public metrics, and exemptions require strict adherence to defined thresholds. Global South initiatives are subject to China’s framework, not independent development pathways.

Source: The Decoder