Proposed rules would restrict overseas transfers of advanced AI models, chip designs and computing technology, with some registration requirements taking effect immediately.
China’s draft export controls would place some of its most advanced artificial intelligence capabilities under a tighter licensing regime, in a move that underscores how strategic AI has become in Beijing’s policy thinking. On 16 October 2026, the Ministry of Commerce and the Ministry of Science and Technology released proposed revisions to the Catalogue of Technologies Prohibited or Restricted from Export, a long-standing tool that China uses to control the transfer of sensitive technologies abroad. According to Bloomberg, the consultation runs until 15 November and covers advanced AI systems, chip design and computing clusters, with the draft also touching model weights and source code for products that exceed certain performance thresholds. AP reported that the government framed the proposal around national security concerns.
The scale of the proposal is unusually broad. White & Case said the draft covers 40 technologies across artificial intelligence, biopharma and photovoltaics, with 23 of those items related to AI. The AI list includes large language models, agents, chip design and computing-cluster technology, while MLLLM said 14 items are proposed for a full ban and 26 for export restrictions. The catalogue is not a new law, but a revision to an existing export-control framework, which means Beijing is extending an older regulatory model to technologies that now sit at the centre of industrial competition.
The most closely watched threshold is the proposed treatment of very large models. Bloomberg said the draft would restrict model weights and source code for products above certain performance levels, while AP reported that models with more than two trillion parameters, or with world-leading capabilities, would come under scrutiny. That dual test matters because it combines a size threshold with a capability threshold. In practice, it gives regulators two routes to capture systems that they regard as strategically important, even if a model does not meet both conditions.
The draft also appears to treat training data and deployment context as part of the control framework. AP said the proposal includes systems trained with sensitive national data, and the more detailed summaries indicate that models built on China’s core data or important data face different levels of restriction. That approach would make a model’s training history a compliance issue, not just its code or architecture. It would also raise the burden on developers that rely on large, mixed training pipelines, because they would need to document where data came from and how it was classified.
Another notable feature is that the draft reaches beyond models themselves into the surrounding infrastructure. Bloomberg said the proposal covers computing clusters, and the other summaries point to controls on AI-chip design and advanced packaging. That is important because frontier AI depends on far more than model code alone: it also depends on the chips, interconnects, orchestration software and packaging techniques that make large-scale training possible. By including those layers, Beijing is trying to control not only finished products, but also the know-how needed to reproduce them.
The draft is also notable for what it does immediately. White & Case said interim registration requirements took effect on the date of the announcement, meaning affected transfers are already subject to a regulatory gate while the consultation is under way. The summaries describe that requirement as applying to transfers of ownership abroad, overseas licensing, and changes that place ownership or actual control in foreign hands. That means companies do not need to wait for the final catalogue text before assessing whether a transaction may need to be registered.
Despite the tougher posture, the reports suggest that ordinary overseas use of Chinese AI may continue through websites and cloud services. One summary says API access and app-based access are excluded, while other coverage says the policy is aimed more at transferring the underlying technology than at blocking every foreign user. That distinction matters for companies outside China that rely on Chinese models through hosted services rather than self-hosting weights. It also suggests Beijing is trying to preserve international reach for its AI sector while limiting the export of the most valuable technical assets.
The wider context is a sharpening technology contest between China and the US. AP noted that Washington still maintains restrictions on China’s access to advanced chips and semiconductor manufacturing equipment, and said Beijing’s move reflects a growing fragmentation of technology trade. China’s draft appears to mirror that pressure by protecting its own AI stack from easy outbound transfer. The consultation remains open until 15 November, so the final scope could still change, but the direction is clear: Beijing is treating advanced AI not merely as a commercial sector, but as strategic infrastructure.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
