The 2026 World AI Conference produced a signature, a speech — and paperwork. On 16 July, 29 countries signed the convention establishing the World AI Cooperation Organization (WAICO). On 17 July, Xi Jinping delivered the keynote we covered from the full transcript. And on the same day as the speech, two action plans were released into the summit's slipstream: the National Development and Reform Commission's AI Cooperation and Development Action Plan (《人工智能合作发展行动计划》) and an International AI Ethics Governance Action Plan (《国际人工智能伦理治理行动计划》) associated with the Ministry of Industry and Information Technology.

We went looking for the primary texts. Here is what we found — including the part where one of them can't be found.

The NDRC plan: a thousand characters, eight lanes

The NDRC published the cooperation plan's full text on its own site on 17 July, attributed to its Department of High-Tech and issued "together with relevant departments" — which departments, it doesn't say. The entire document is eight numbered "actions" totalling roughly a thousand Chinese characters. That is not a summary of the document. That is the document.

The eight actions, compressed (all English renderings in this piece are our translations of the Chinese text):

1. Quality data supply — advance cross-border data flows, build "trusted cross-border data spaces" in selected sectors, jointly build multilingual corpora.
2. Universal-access compute — provide "inclusive AI-compute services" to developing countries; jointly build green-energy-driven infrastructure.
3. Open-source ecosystem sharing — share general-purpose large models, base algorithms and tooling; "jointly set open-source compliance systems and safety norms"; support countries doing "localised innovation on open models."
4. Deep AI enablement — "AI+" cooperation across science, manufacturing, healthcare, education, agriculture and governance. Notably, the text singles out AI agents (智能体) by name, calling for their "regulated application and innovative development."
5. Joint talent cultivation — joint training mechanisms, occupational standards and skills certification, and a line on "jointly addressing the impact of AI's rapid development on employment."
6. Rules and standards — a joint AI standards system, pushing international standards drafting and revision, "inclusiveness and interoperability."
7. Safety governance collaboration — cybersecurity threat-intelligence sharing, preventing "misuse and abuse," research into explainability, transparency and safety.
8. AI for good — ethics principles, "eliminate racism, discrimination and other forms of algorithmic bias," public scientific goods for the Global South, in service of the UN 2030 Sustainable Development Goals.

What's not in it

Numbers. The plan has none of its own — the only digits in the text are in the name of the UN's 2030 Agenda. No budgets, no timelines, no country counts, no programme names. Every verb is aspirational: advance, promote, encourage, support, co-build. And unlike the WAICO convention, which was signed (签署) by 29 founding states the day before, this plan was released (发布) — by China, unilaterally. Nobody else's name is on it.

The concrete figures of WAIC week — 5,000 AI training places, AI cooperation centres for six regional blocs, the "Mazu" weather system offered to 30 countries — all come from Xi's keynote, not from this document. The most useful reading is that the plan is standing scaffolding: the keynote's pledges each slot into one of its eight lanes (training places into talent cultivation, the compute and weather offers into universal-access compute and AI-for-good), while the plan itself commits to nothing countable. Its stated purpose, per the NDRC's companion release, is to implement Xi's formulation that AI should be "an international public good benefiting all humanity" — the same phrase doing the same work across the whole summit.

One housekeeping note, because English-language summaries are already mixing them up: this is a different document from July 2025's 13-point Global AI Governance Action Plan (《人工智能全球治理行动计划》), whose full text sits on gov.cn. Same genre, same city, different year, different document.

The ethics plan you can't read yet

The second document is stranger. The International AI Ethics Governance Action Plan was announced on 17 July in a three-paragraph Xinhua wire — and as of publication, that wire is effectively all there is. The full text is not on gov.cn, not on the MIIT's site, not on the foreign ministry's. Every outlet carrying the story, including Beijing News, runs the identical wire copy.

The wire's own phrasing is precise in an interesting way: MIIT "together with relevant parties jointly guided the release" (会同有关方面共同指导发布) of the plan. Guided — not issued. Who formally issued it is not stated anywhere we can find.

What the wire does say: the plan is framed as "an important public good China offers the international community," placed explicitly under the UN's Pact for the Future and its Global Digital Compact annex, and organised around directions including full-lifecycle ethics governance, classified and tiered risk control (风险分类分级防控), agile governance mechanisms, collaborative industry-chain ecosystem cultivation, and fostering a supportive social atmosphere. That second item is the line worth underlining: risk classification and tiering is the taxonomy at the heart of China's domestic AI governance framework — the 2023 generative-AI Interim Measures mandate "inclusive, prudent, classified and tiered regulation" — and this plan proposes it as an international organising principle. The follow-through list includes ethics-governance capacity building aimed at developing countries, open-sourcing of explainability, privacy-protection and bias-correction tooling, and protections for women, children, the elderly and disabled people.

Is it binding? No — the wire says the plan "calls on countries to act in light of their own circumstances." Are there numbers? None reported.

There is precedent for the missing text turning up: 2025's governance action plan got its full text posted on gov.cn the same day as its conference debut. When this one's primary appears, we'll read it and update this piece in place.

The pattern

Read together, the paperwork tells one consistent story. The binding thing (WAICO, signed by 29) is institutional; the programmatic things (two action plans, released by China alone) are offers — compute, corpora, open models, training, ethics frameworks — aimed squarely at the Global South, wrapped in UN framing, with standards-setting and open-source governance as the lanes where China is bidding to be convener. The numbers live in the speech; the speech's scaffolding lives in the plans; and the most consequential document of the three may be the one whose text nobody outside the room has read.

We'll add the ethics plan's primary text when it surfaces, and we're watching for WAICO's full founding-membership list and any document-specific international reaction — day-one Western coverage focused on the keynote and the signing.