China's AI week: a governance agenda from the NDRC, DeepSeek's record round and a safety turn at Wuzhen

China moved to formalize the rules of its artificial intelligence boom this week, pairing a sweeping regulatory agenda from its top economic planner with a fresh wave of capital and model releases from the country's leading AI labs.

At a press conference on Nov. 3, the National Development and Reform Commission set out an AI governance agenda for the 15th Five-Year Plan period, commitments that Trivium China, a Beijing-based policy consultancy, described as "the most comprehensive single statement on AI regulation" yet issued by a national-level agency. The plan signals that Beijing intends to keep AI development inside a defined regulatory framework.

The push came from several directions at once. On Nov. 2, Zhuang Rongwen, director of the Cyberspace Administration of China, published an essay acknowledging the "double-edged nature" of AI and other new technologies, according to Trivium's translation and analysis. A day later, Premier Li Qiang used his keynote at the World Internet Conference in Wuzhen to argue that innovation and safety must advance together, according to state media coverage via Trivium China. The industry ministry, meanwhile, published the fifth batch of its 2026 industry-standards revision programme on Nov. 3 — granular technical rulemaking that runs beneath the headline agenda, indexed with its official text in the Caybles tracker. And regulators reached for less conventional tools as well: according to Caixin's reporting, translated by Trivium China, authorities moved against the "black-hat" search-optimisation industry and its data-poisoned advertisements.

While officials emphasized caution, the industry's appetite for risk appeared undiminished. DeepSeek, the Hangzhou-based lab that startled Silicon Valley with its low-cost models, closed a second funding round of about 105.5 billion yuan ($15.8 billion) from nearly 30 investors — the largest venture round ever for a Chinese AI firm — with state-backed investors supplying about 27 percent of the capital at a valuation of roughly 615.5 billion yuan ($92 billion), according to Trivium China's deal tracking.

Tencent, meanwhile, released HY4, an open-weight model that filters offensive cyber and biological data from its training set. "Is this the new standard for strong Chinese open-source model releases going forward?" wrote Kyle Chan, a researcher at Princeton University who studies Chinese technology policy, on X. "Don't count Tencent out of the LLM model race in China," wrote Kaiser Kuo, host of the Sinica Podcast, calling the release's emphasis on safety "good to see." Chan also flagged DeepSeek's V5 running on Huawei chips and Xiaomi's humanoid-robot push, with its own vision-language-action model, as milestones from the same run of releases.

That question hung over Wuzhen, where researchers from DeepSeek, Alibaba's Qwen and MiniMax discussed open-source AI on a panel on Nov. 3. Executives from DeepSeek and Qwen conceded that open-weight models carry inherent risks — a notable admission from labs whose products are distributed freely, and a marker of the safety turn we tracked earlier this autumn. The summit this year was built around open-source technology and international cooperation, according to Trivium's conference coverage.

The optimism was not universal. Speaking at the International Financial Leaders Investment Summit in Hong Kong on Nov. 3, Xiao Yuanqi, deputy head of the National Financial Regulatory Administration, warned that the global rush into AI is crowding out investment in the rest of the economy, a rare cautionary note from a financial regulator, as Trivium China reported. It was not an isolated note: the securities regulator imposed a share-buyback mandate on some listed firms a day earlier, according to Trivium's market-policy tracking, and the central bank raised its foreign-exchange reserve ratio on Nov. 3 to cool the renminbi's rally, per its central-bank watch. Not everyone read the caution as bearish — Adam Tooze, the economic historian, argued that anti-AI regulation politics is "actually GOOD for AI."

Markets, at least, were unbothered. The SSE STAR 50, home to many of China's chip and AI listings, jumped 9.9 percent on Nov. 4, while the Hang Seng Tech index rose 1.8 percent, according to the Caybles market tracker.

Whether Beijing's regulatory ambition can keep pace with its labs' output remains an open question. For now, the message from Wuzhen was clear enough: China wants to be seen as both an AI innovator and a responsible one.

Further coverage

• Bill Bishop, Sinocism — Li Qiang on AI, Nvidia-China, and skepticism about the latest rescue
• Bill Bishop, Sinocism — the Nvidia chip and export-control chain
• Trivium China × Sinica Podcast — China's critical minerals and "managed interdependence"
• MIIT — monitoring-chemicals export-control compliance guidelines (Oct. 29 consultation), official notice at miit.gov.cn

Tracking the underlying documents: every policy referenced above — including the NDRC statement, the CAC essay and the MIIT standards notice — is indexed in the Caybles policy tracker with a link to its official Chinese original.