China proposes export controls on AI models, chips and agents

By Splitfeed EditorialLast updated Oct 16, 2026, 10:40 UTC

On October 16, 2026, China formally proposed adding approximately 40 technologies to its export control list. The proposed catalogue covers AI models, AI chips, AI agents and model weights, alongside biotechnology and solar power technologies. While full restrictions remain subject to public comment, China has already imposed temporary export registration requirements on the listed items. The same day, Taiwan's minister Kung Ming-hsin confirmed that Taiwan intends to expand its own export controls and introduce criminal penalties for anyone caught smuggling AI chips into China.

The move is significant because it marks China's most explicit attempt to restrict outbound flows of AI technology, including not just hardware like chips but also software artifacts such as model weights and autonomous agents. Until now, export control friction in the AI sector had been largely asymmetric, with the United States and allied governments restricting what could flow into China. A Chinese regime that mirrors those controls changes the geometry of the technology competition, potentially limiting how freely Chinese AI models and chips can reach third-country markets, research partners, or companies that have come to rely on Chinese supply chains. What remains genuinely in dispute is how strictly these controls would be enforced, which trading partners would be most affected, and whether the proposal is primarily a negotiating instrument or the foundation of a durable new policy.

Because no reactions from the Pro-AI, Anti-AI, or Middle Ground camps have been published yet, their likely positions can only be characterised in general terms. The Pro-AI camp would typically argue that export restrictions on AI models and agents risk fragmenting global research collaboration and slowing the diffusion of beneficial technology, and might frame the controls as a net harm to innovation on both sides. The Anti-AI camp would generally welcome any measure that slows the spread of powerful, poorly governed AI systems across borders, and might see mutual controls as a rare case of governments exercising meaningful oversight. The Middle Ground camp would likely focus on the practical difficulty of controlling software assets like model weights, and call for multilateral frameworks rather than unilateral or tit-for-tat restrictions.

The public comment period on the full restrictions is the immediate thing to watch, since the final catalogue, its definitions and its enforcement mechanisms will determine whether the proposal has real teeth. Taiwan's legislative timeline for enshrining criminal penalties is a parallel development worth tracking, as is any response from the United States or the European Union signalling whether they view China's move as an escalation, a form of retaliation, or an opening for coordinated multilateral controls.