China's NDRC moves to quicken lifecycle AI safety legislation

China's state economic planner, the National Development and Reform Commission (NDRC), said on November 3, 2026 that it is quickening legislation to govern and oversee how AI is developed and applied, covering the entire life cycle of models. The move would create a broad set of standards described as representing some of the most comprehensive guidelines to guard against a loss of control over increasingly powerful artificial intelligence.

What it is

The planned instrument is new AI safety legislation led by China's state economic planning agency rather than its internet or cyberspace regulator, aimed at overseeing artificial intelligence from the development stage through deployment. It is being pursued as a single, life-cycle-spanning set of standards rather than a narrower measure covering one stage such as pre-release review.

What it requires

According to Wang Ruomeng, a senior NDRC official, the regulations will extend to areas from data to computing power.

The scope encompasses developers from DeepSeek to Moonshot and Alibaba Group Holding Ltd., who are leading a nationwide effort to create models that can hold their own against the best from OpenAI and Anthropic.

What security leaders should do

Security and compliance teams working with Chinese frontier-model providers should track the drafting process closely, since the planned rules are described as covering the entire life cycle of models rather than a narrow slice such as pre-deployment review. Teams should also monitor how the eventual standards treat computing power and data, the two areas the NDRC specifically flagged, because those requirements could affect sourcing and infrastructure decisions involving vendors such as DeepSeek, Moonshot and Alibaba. Because the initiative is still described as legislation being quickened rather than finalized text, security leaders should treat any current compliance assumptions as provisional and revisit this entry once draft rule text is published.