AI firms face new export rules as China targets advanced models and chip blueprints for control
New draft rules would restrict the overseas transfer of advanced AI models and semiconductor blueprints, though open-source software remains exempt from these measures.
Key points
- China’s ministries proposed adding 23 AI technologies to an export control list, covering large models and chip designs.
- The draft exempts open-source code and allows overseas access via apps, while requiring registration for exports.
- Rules target models with over 2 trillion parameters or those ranking in the global top 10 on performance tests.
China’s commerce and science ministries have proposed a draft regulation that would place 23 artificial intelligence technologies under strict export controls. This move aims to restrict the international transfer of advanced AI capabilities, including large language models, autonomous agents, and specific semiconductor designs. The proposal seeks to manage how these powerful tools leave the country, marking a significant shift in how Beijing regulates its technological exports to foreign entities and markets.
The draft proposal, released on a Friday, covers a broad spectrum of high-end technologies. According to reports from Bloomberg and The Next Web, the controls apply to AI models with at least 2 trillion parameters or those that rank among the world’s top 10 in performance and safety assessments. This effectively targets some of the most capable systems currently available, including models developed by major Chinese tech firms, although the document itself does not name any specific products or companies.
What we know
The proposal encompasses 40 items in total, extending beyond AI to include 12 solar energy items and five biomedicine items, such as gene therapy. For AI chips, the restrictions cover design know-how for semiconductors manufactured at 14 nanometres or below, as well as those built using 2.5D or 3D stacking techniques. Some memory chip designs are also included in this list. Firms exporting these items must now register under a temporary rule that took effect on 16 October, which also applies to incomplete deals.
A ministry spokesperson stated that the rules aim to prevent misuse and ensure AI remains under human control, adding that the plan does not target any specific country or firm. The public can submit comments on the draft until 15 November. Notably, open-source code and published research are exempt from these controls. Chinese companies can continue to serve overseas users through applications, websites, and APIs, meaning end-users abroad can still access these models without restriction.
Running a covered model on servers located outside China would count as an export under the new rules. However, firms that operate these models on their own cloud servers abroad may receive an automatic licence. This distinction suggests a focus on the transfer of proprietary technology rather than the mere availability of services. Earlier this year, reports indicated Beijing was weighing such limits, with the Financial Times noting considerations in July regarding export controls on AI models and chips.
Defender's checklist
- Monitor official channels for finalised regulations and implementation timelines.
- Review current supply chains for dependencies on Chinese AI models or chip designs.
- Audit software portfolios to distinguish between proprietary and open-source components.
- Assess compliance requirements for any existing cross-border data or model deployments.
The background in plain language
Export controls are government restrictions on the transfer of sensitive technologies to foreign entities. They aim to prevent strategic advantages from shifting to competitors or adversaries. For AI, this often involves limiting access to highly capable models or the hardware needed to run them. Teams usually assess their technology stack to identify any components that might fall under such lists. Understanding these rules helps organisations manage legal risks and ensure they remain compliant with international trade laws, avoiding potential penalties or service disruptions.
Background: Open-source software
Open-source software is code released with permissions that allow anyone to view, modify, and distribute it. This transparency enables community-driven development and rigorous peer review, distinguishing it fundamentally from proprietary systems where the internal logic remains hidden from the user.
Read the full guide: Open-Source Software: Meaning, Origins and Daily Use
What to do and how to stay safe: AI export controls
- Review contracts with Chinese AI vendors to understand data residency and export clauses.
- Verify if your current AI models are open-source or proprietary to determine compliance needs.
- Consult legal experts to interpret how the new registration rules apply to your operations.
- Monitor for official updates once the vendor provides an update on the final regulations.
Step-by-step guide: AI Governance for Small Teams: Practical Controls and Cost
General security guidance from the TechWire Beat newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
Does this ban affect open-source AI models?
No, the draft proposal explicitly exempts open-source code and published research from the export controls.
Can overseas users still access Chinese AI models?
Yes, users abroad can still access models through apps, websites, and APIs, as these services are not restricted.
What happens if a company runs a covered model on foreign servers?
Running a covered model on servers abroad counts as an export, though firms using their own cloud servers may get an automatic licence.
