On November 3, China's National Development and Reform Commission said at a press conference in its 14th Five-Year Plan series that it will accelerate foundational and comprehensive legislation on artificial intelligence, completing safety standards and ethical guidelines covering the full AI lifecycle. Wang Ruomeng, deputy director of the commission's Department of Innovation and High-Tech Development, named DeepSeek, Moonshot and Alibaba, saying legislation and oversight will cover key areas such as data, computing power and algorithms, with the aim of preventing increasingly capable AI from slipping out of control.
This was not a law already passed, but a signal that legislation is accelerating. The work priorities Wang listed include completing the governance rule system, improving the regulatory environment, and clarifying the shared governance responsibilities of government, industry organizations, enterprises and the public. The core idea is to make development and security work together inside enforceable rules, with safety running through research, application and governance, regular assessments of implementation, and revisions to provisions that fall behind technology.
What will be covered: from data and compute to the full lifecycle
According to the press conference, legislation will complete laws for common elements such as data, computing power and algorithms, add dedicated rules for key application scenarios, and establish safety standards and ethical guidelines covering the full lifecycle. Developers and operators will need to disclose safety risks promptly and report security incidents and vulnerabilities. For the leading firms named, that means a model release is no longer only a capability contest: risk disclosure and compliance records will become fixed costs.
What buffers remain
Regulation will not arrive in a single cut. The commission proposed reasonable transition periods for policies, compliance guidance by industry, support for AI regulatory sandboxes, and improved mechanisms for application, testing, use and exit. Regular assessment and sandboxes run in parallel, leaving room to trial the new rules: verify them in limited scenarios first, then roll them out industry-wide. For smaller firms, the real unknowns are how long transition periods will last and what sandbox entry conditions will be — details the press conference did not disclose.
What comes next is the legislative text and timetable. The commission named no draft title or deliberation dates, so compliance costs and disclosure duties for leading firms can only be tallied once detailed rules land. Regulators in China and elsewhere are moving from after-the-fact accountability to lines drawn in advance: California's AB 1883 bars employers from using AI to infer workers' emotions from 2027. The pace of corporate compliance is increasingly set by legislative timetables.
