Beijing Says It Will Push AI Developers to Report Their Own Incidents and Flaws

China wants the people who build AI models to tell the government when those models go wrong. On Tuesday, the state economic planner said it is speeding up foundational, comprehensive AI legislation, and the most interesting piece is a push for developers and operators to disclose safety risks and report their own incidents and vulnerabilities, Bloomberg reported.

Nobody offered a draft text, a vote or a compliance deadline. The National Development and Reform Commission (NDRC) offered a direction of travel, delivered in a press conference series on the next Five-Year Plan.

The speaker was Wang Ruomeng, deputy director-general of the NDRC's Department of Innovation and High-Tech Development. According to the Beijing Daily's account of the briefing, published in Chinese, Wang said there is now broad consensus that AI brings opportunity and risk together. Security, Wang said, will run through research, deployment and governance in the coming plan period.

"AI safety is not a problem one company, industry or country can tackle alone" Wang Ruomeng, NDRC, as quoted by Bloomberg

Bloomberg reports that Wang paired that line with a call for countries to coordinate on vulnerability disclosure and risk responses.

A duty to confess, softly worded

Here is where reading the original matters. Bloomberg and the Korean broadcaster SBS, which cited Bloomberg, say developers and operators must or should disclose risks, incidents and vulnerabilities. The Beijing Daily's Chinese text is gentler. It says the government will push developers and operators to strengthen risk awareness, disclose safety risks promptly and proactively report security incidents and vulnerability information.

Pushing is not mandating. Whether this becomes a hard obligation with penalties is for the eventual law to settle. For now it is a stated intention.

The rest of the package leans toward carrots. Wang said the government will set a transition period for regulatory policies, though the length was not specified, and will give industry-specific compliance guidance. The Beijing Daily says officials will also build assessment, consultation and certification services to lower compliance costs, and back sector-specific regulatory sandboxes with procedures for applying, testing, using and exiting them. Rules will be reviewed periodically and outdated provisions revised.

That mix is deliberate. Per Bloomberg, Beijing is trying to balance the risk of AI running amok or fuelling cyberattacks against competing with the US. Bloomberg also reported that the government will keep encouraging open AI development through the 15th Five-Year Plan, which ends in 2030. The open-model strategy is not being put on a leash.

The labs in the frame

Bloomberg says the rules would cover developers from DeepSeek to Moonshot and Alibaba Group, which are leading China's push to build models that can match the best from OpenAI and Anthropic. SBS likewise says those firms are expected to be included. Treat that as Bloomberg's reading. None of the sources I reviewed show an official list of companies.

Wang's list of social risks goes well beyond usual lab-safety talk. According to the Beijing Daily, authorities will watch how AI changes employment, strengthen monitoring and early warning on job replacement, and tie the social security system to changing work patterns. New occupations such as model trainers and data annotators are to be added to the national occupational classification. The briefing also flagged algorithmic discrimination, deepfakes, privacy leaks, protection for older people and minors, and guidance against addiction and emotional dependence on AI.

Job losses are a live fear elsewhere too. Goldman Sachs has said AI could push 1 in 11 workers out of their jobs. A planner promising a safety net before the layoffs arrive is at least a different opening position.

The timing has a diplomatic edge. SBS, citing Bloomberg, notes that Donald Trump and Xi Jinping agreed at their September summit in Washington to exchange views on AI risks and to set up a communication channel for AI accidents. An incident-reporting culture at home would give such a channel something to carry. That is my inference, not the NDRC's.

The skeptic's question is simple. Who receives the reports, who decides what counts as an incident, and what happens to a lab that stays quiet? Until a draft appears, nobody knows. The announcement is real. The law is still a promise.