China's draft export catalog covers 23 AI technology areas and auto-licenses open-weight releases
China's commerce and science ministries drafted export controls on 23 AI technology areas, with model limits from 2 trillion parameters or 10^26 FLOPs. A top-10 global rank also counts, open-weight releases clear on registration, and public technology is exempt.
The Engineer · Build desk
Key figures from China's draft export catalog. Controls on a large model's core technology apply once it has at least 2 trillion parameters, hits 10^26 FLOPs of training compute, or ranks in the global top 10.
| Figure | Value | Note | Claim |
|---|---|---|---|
| Model parameters, at least | 2 trillion | 4 | |
| Training compute, power of 10 | 26 10^x FLOPs | Threshold is 10^26 FLOPs | 4 |
| Global ranking, top | 10 rank | 4 |
What happened
- AI tools that write or alter malware on their own to attack critical infrastructure, the military or government face a total export ban, and some vulnerability-finding tools get new limits.
- Once a model crosses a threshold, the controls reach its weights, checkpoints, source code, architecture and training or inference methods.
- A separate trigger covers models trained on China's "important data"; lawfully public data is exempt, and up to 0.1 percent is allowed if three conditions are met.
- API calls, apps, SaaS tools and lightweight on-device versions of restricted models stay outside the controls, but installing one on local servers abroad needs a standard export license.
- In the meantime, before the final rules apply, affected exports have to be filed on the commerce ministry's portal.
Why it matters
- What it limits constraint A developer cannot settle its top-10 status from its own training run, because rank moves when other labs release and officials choose the rankings.
- What's now possible capability Chinese firms with overseas data centers, chip tape-outs or foreign research partners can ship restricted tech on filing, without waiting on a license decision.
- What doesn't add up contradiction The Decoder expects the draft to push US firms toward domestic models and weaken Europe's open-weight strategy, yet the draft it describes licenses open-source releases automatically.
Two of the three model thresholds are properties of the model and its training run, so a lab can check them itself [4]. The third is judged by someone else. Export officials decide top-10 status from public rankings by major domestic and foreign evaluators or industry platforms, and they can look at overall performance, specific tasks, safety or global usage numbers [5]. The draft as reported leaves the choice of ranking and measure to the officials [5].
The data trigger is harder to certify than it looks. The 0.1 percent allowance has three conditions: the data was cleaned beforehand, it never shows up in AI outputs, and it does not substantially boost the model's R&D capacity for core tech in that field [6]. The first and third can be documented. The second is a claim about every output the model will ever produce.
The exemptions are layered. AI tech already public through open-source releases or research papers skips the rules entirely [10]. New open-source publication of restricted tech and weights gets automatic licensing, and eligible exports clear the moment the filing is made [9]. Software that runs on a company's own overseas cloud can use the streamlined process too [8]. The Decoder says the fast track exists to avoid choking off China's domestic open-source boom [13].
That leaves the standard license for a narrow case. We think the friction falls on closed weights shipped for self-hosting abroad [7]. Open-weight publishers face only a filing [9]. The Decoder's report does not give an effective date for the final rules.
What to watch
- The final text: whether the thresholds, the 0.1 percent data allowance and the automatic open-weight license survive unchanged.
- Which rankings and evaluators export officials name for the top-10 test.
- Whether the standard license for local-server installs abroad stays as drafted.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence48
- Adoption
- Not scored
- Hype gap
- Not scored
- Incentives
- Not scored
- Confidence45
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
China's commerce and science ministries have released a draft update to the nation's list of banned and restricted export technologies.
- [2]
The catalog targets 23 key AI technologies across large models, agents, compute clusters, and data processing.
- [3]
AI tools that can write or alter malware on their own to attack critical infrastructure, the military, or government agencies face a total export ban; certain AI tools for finding software vulnerabilities will not be banned outright but will face new export limits.
- [4]
Controls on the core technologies behind large models (weights, checkpoints, source code, architecture, and training or inference methods) kick in once a model hits 10^26 FLOPs of training compute, has at least 2 trillion parameters, or ranks in the global top 10.
- [5]
Export officials determine top-10 status using public rankings from major domestic and foreign evaluators or industry platforms, looking at overall performance, specific tasks, safety, or global usage numbers.
- [6]
A separate case covers large models trained on China's 'important data' (as defined by the Data Security Law and related rules) that offer services based on it. Lawfully public data is exempt. Up to 0.1 percent is allowed if the data was cleaned beforehand, never shows up in AI outputs, and does not substantially boost the model's R&D capacity for core tech in that field.
- [7]
Using restricted models through API calls, apps, websites, SaaS tools, or lightweight versions on local devices stays free of these export controls; installing restricted AI tech on local servers abroad requires a standard export license.
- [8]
Companies can still use the streamlined process if the software runs on their own overseas cloud.
- [9]
Publishing restricted AI tech and weights as open source qualifies for automatic licensing, as do Chinese firms that run their own data centers abroad, tape out chips overseas, conduct research with foreign partners, or help set international standards; eligible exports are cleared the moment a company files the paperwork.
- [10]
AI tech that is already public through open-source releases or research papers is exempt from the new rules.
- [11]
Until the final rules take effect, companies must register affected exports through the commerce ministry's portal.
- [12]
The Decoder says the draft will likely push US companies that rely on open-weight AI to double down on domestic models, and argues that the EU's strategy of running open-weight models in domestic clouds falls apart when the best open models come from the US or China and can be cut off by export controls.
- [13]
According to The Decoder, China is introducing the fast-track process to avoid choking off its domestic open-source boom.
Get this ledger inside ChatGPT. Ask about any of our stories and the answer brings its figure, each claim’s verdict and the sources. Free, no account.
Sources
2 independent publishers whose own reporting we read for this story.
- mezha.netКитай пропонує додати 40 технологій ШІ, біомедицини та сонячної енергетики до експортного контролю
1 article · October 16, 2026
- runtimewire.comChina proposes export controls on 40 technologies across AI, biotech and solar
1 article · October 16, 2026
- the-decoder.comChina tightens export controls on open-weight models and 23 key AI technologies
2 articles · October 16, 2026
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.
Entities
- The DecoderFollow
- China's Ministry of CommerceFollow
- World Artificial Intelligence Cooperation OrganizationFollow