China now requires firms to register transfers of 23 AI technologies
The final export catalogue is still a proposal, but pre-export registration took effect on 16 October. APIs and ordinary app access are not automatically caught.
The short version
3 key points
- 1.China opened consultation on adding 40 technologies to its prohibited or restricted export catalogue, including 23 covering large models, agents, computing clusters and information processing.
- 2.A temporary requirement to register covered technology exports began immediately on 16 October and includes transfers already under way but not completed.
- 3.The official explanation says API, app, web and SaaS access to restricted large models is not itself a technology export; overseas local deployment can be in scope when restricted know-how is transferred.
The immediate change is registration, not a final blanket ban
Chinese firms transferring certain AI know-how abroad must now register the transaction before export. The requirement took effect on 16 October and applies while China consults on a wider revision of its prohibited and restricted technology catalogue. The draft covers 40 technologies: 23 in artificial intelligence, five in biomedicine and 12 in solar power. Across all three fields, officials propose prohibiting 14 and restricting 26.
That distinction matters. The final catalogue is not yet settled, and consultation remains open until 15 November. Registration is an immediate information duty, not proof that an export has been refused. The announcement says it covers ownership transfers, licences for overseas use, a change in the technology owner or controller to a foreign entity, and other cases specifically notified by the ministry. Deals already started but not completed are included. Filings identify the exporter, recipient, destination, technology, intended use and expected value; multiple technologies going to one recipient can be combined.[1][2]
Model access and technology transfer are treated differently
The official explanation narrows the practical reach. It says most AI technology can still be exported freely. Material already placed in the public domain through open source or academic publication is excluded. It also says providing a restricted large model through an API, app, website, SaaS product or lightweight on-device deployment is not, by itself, treated as exporting the underlying model technology.
Overseas local deployment is different when it moves restricted know-how out of China. Deployment to a company’s own overseas cloud may qualify for an automatic-licence route, as may overseas chip fabrication, joint research using foreign resources and some international standards work. The draft’s cited AI examples include a proposed prohibition on technology that autonomously generates and mutates malicious code, and a proposed restriction on AI-assisted vulnerability discovery. Products made with a listed technology are generally separate from the transfer of the technology itself.[1][2]
The distinction that matters
Three routes under the 16 October policy
The policy turns on whether a transaction transfers listed technical capability, not merely whether somebody outside China can use an AI service.
Service access
Offer a model through an API, app, website or SaaS
Not treated as a large-model technology export in the official explanation
Technology transfer
Transfer ownership, license overseas use or move control abroad
Register before export now; licensing depends on the final category and route
Public or facilitated
Use already-public material or an eligible own-cloud, research or standards route
Public-domain material is excluded; specified cases may receive automatic licensing
Companies need a transaction-level check now
For developers and their overseas partners, the operational question is no longer simply whether a model can be accessed abroad. They must identify whether a contract, licence, deployment or ownership change transfers controlled technical capability, then determine whether registration, an automatic licence or a full export licence applies. That can affect deal timetables, technical documentation and how cross-border teams separate hosted services from transferable know-how.
The evidence is an official policy announcement and its same-day interpretation, not data on enforcement or economic effect. The ministries publish no denominator for current technology exports, expected registrations, approval rates or compliance cost. The draft can change after consultation, and detailed thresholds in its attachments will determine which systems are actually caught. The assessment will change when the final catalogue and implementing guidance show the precise technical tests, licence pathways, review times and treatment of existing contracts. Until then, claims that China has banned every AI model export go beyond the text.[1][2]
Evidence check
The sources describe the government’s policy and interpretation; they provide no enforcement record, registration denominator, approval rate or measured effect on firms. The final catalogue may differ from the consultation draft.
Reporting basis: Source analysis. Original source date: 16 October 2026.
Explore the full source and evidence record
What this means for people
- Engineers and legal teams may need to separate hosted access from transfers of model know-how before signing or continuing overseas arrangements.
- Foreign partners may face extra documentation or delay even where a service remains available through an API or SaaS interface.
Global context
The measure governs technology leaving China and can affect partners worldwide. It does not target a named country or entity, according to the ministry, and it should not be read as a global ban on Chinese AI products.
What the evidence does not yet show
- The sources describe the government’s policy and interpretation; they provide no enforcement record, registration denominator, approval rate or measured effect on firms. The final catalogue may differ from the consultation draft.
What to watch next
- The final catalogue after 15 November, exact technical thresholds, guidance for existing contracts, and published evidence on registrations, automatic licences and refusals.
Living evidence record
Impact record IAI-0669S1R
Evidence stage
Announced
Confidence
Supported
Reporting basis
Source analysis
Independent or research support
Not yet
Record status
Monitoring
Last checked
16 October 2026
Source trail
2 direct sources across 1 source type.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Related-source reporting disclosure
This record analyses 2 linked source records around the same underlying development. The extra records add method, date or context, but they do not by themselves constitute independent replication of every performance claim or predicted outcome.
Evidence trail
Sources used for this report
Links checked 16 October 2026
Research question, topics and regions
Topics: AI exports · Technology transfer · Export controls · AI agents · Cybersecurity
Regions: China · Global
This report is labelled source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
The Impact Brief · Free
Keep following the stories that matter to you.
Get our weekday briefing on what changed and why it matters. Choose your topics below.
Continue the story
There’s more to this story
Government & Policy
Will AI answer engines narrow the open web?
They could, Europe’s telecom regulators warn: direct answers and agents may reduce source diversity, weaken publisher incentives and shift choice toward a few gateways. But this is a draft evidence review for consultation—not a final rule or a new causal study.
9 min · 3 sources
Government & Policy
What must Australian governments prove before using AI?
A national framework published on 16 October asks federal, state and territory bodies to document purpose, assess impacts, govern data, test systems, monitor performance and keep humans accountable. It is a shared assurance baseline—not proof that every agency has implemented those controls.
7 min · 4 sources
Government & Policy
Can medical-device AI assessments be reproduced?
Fourteen hosted model deployments changed at least one repeated assessment despite low-variance settings. A company-built pipeline reproduced every stage on fixed hardware, but the study tested repeatability—not correctness, usefulness or regulatory approval.
9 min · 1 source
Reader commentary
What does this mean where you are?
Tell us what you have seen at work, in your community or in everyday life. A question is welcome too. No account is needed; we review notes before publishing them.
Explore commentary across the portal →Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.