China is redrawing the boundary for AI technology exports. On October 16, 2026, the Ministry of Commerce and the Ministry of Science and Technology issued Announcement No. 48 of 2026, opening public consultation on a revision of the Catalogue of Technologies Prohibited or Restricted from Export. Comments run from October 16 to November 15, and until the revised catalogue takes effect, exports of the listed technologies will be subject to pre-export registration.
23 of the 40 proposed items are AI
The proposed changes span artificial intelligence, biomedicine and photovoltaics, 40 items in total, with 14 proposed for prohibition and 26 for restriction. AI alone accounts for 23 items, covering large models, agents, compute clusters and information processing. The examples named in the spokesperson's briefing concentrate on security-sensitive capabilities, such as AI-driven generation and mutation of malicious code and AI-assisted vulnerability discovery. The controls target the cross-border transfer of the technology itself; products made with those technologies are not in scope, and most listed items carry specific technical thresholds rather than blanket product bans.
A second carve-out matters just as much: technology that has already entered the public domain through open source releases or published papers is proposed to stay outside the controls. The line is drawn around unpublished know-how still held domestically, not around results already shared with the global developer community.
Registration is not approval, but four situations trigger it
The interim registration applies from the day of the announcement. It covers transferring technology ownership abroad, licensing domestic technology for overseas use, a change of the technology's owner or actual controller to a foreign entity, and any other case the ministry later specifies. Registration runs through the ministry's unified business platform, and exports already under way but not yet completed also fall inside the requirement.
Facilitation runs in parallel. Automatic-licensing style arrangements are reserved for cases such as open-sourcing AI technology, building self-operated overseas compute centers, taping out chips abroad, joint research using overseas resources, and participation in international standards work. For a model going overseas, the test the spokesperson offered is whether restricted technology actually moves from inside China to outside it; overseas local deployment and serving model calls to foreign users both have to be checked against that test, whatever the business model is called.
Why now
The stated reasoning has two sides. One is security: capabilities such as autonomously driven malicious-code generation could, in the wrong hands, be turned against critical information infrastructure, defense and core government systems. The other is openness: China says it supports AI dialogue and cooperation on the basis of mutual respect and mutual benefit, with extra facilitation for capacity-building cooperation among members of the World AI Cooperation Organization. The catalogue will also be adjusted dynamically, removing technology that is no longer advanced.
For teams shipping models abroad, the practical effect is that compliance moves earlier: technology licensing, ownership changes and overseas compute plans now need to be checked against the registration rules immediately, not after the final catalogue lands. Scrutiny is tightening on the US side at the same time, as the federal evidence-gathering described in the FTC's demands to OpenAI and Anthropic shows; read together, cross-border flows of frontier technology are being fitted with registration gates on both sides.
