China gives workers notice before AI takes their job
The short answer
China's State Council issued an 18-point plan on AI and employment. It asks employers to run employment-risk assessments on big AI projects, to give affected staff advance notice through an "unemployment buffer period", and to pay for retraining and transfers. It also covers AI use in HR, algorithmic discrimination, and unemployment insurance for displaced workers. It is the first national policy that treats AI job loss as a labour-rights problem rather than a tech problem.
What is actually in the plan
The guidelines, from the General Office of the State Council, run across six areas: creating jobs through AI, making adoption more employment-friendly, improving digital skills, strengthening employment services, protecting labour rights, and monitoring employment risk.
The concrete parts matter more than the framing. Companies introducing AI at scale are encouraged to establish an "unemployment buffer period", telling affected employees in advance and funding retraining, internal transfers or help finding another job. Projects that need market access approval and could significantly affect jobs are expected to carry an employment-risk assessment, with departments coordinating support for displaced workers.
There is also a caution clause for labour-intensive industries: authorities should weigh the scale, intensity and pace of deployment to limit the impact on workers. And a counter-clause: AI should be prioritised where there are labour shortages or hazardous conditions, where it can improve safety.
The rights section is the unusual part
Most national AI plans talk about capability and competitiveness. This one spends its later points on wages, rest, leave and workplace safety, and says the government will regulate how AI is used in human-resource management. It names three specific harms: employment discrimination, fraudulent recruitment, and privacy violations arising from algorithmic hiring.
It also reaches into social security, proposing better coverage for flexible and gig work, and adjustments to income distribution so workers share in the gains from AI. Whether any of that is enforceable is a separate question. As policy language, it is a clear statement that the gains are not assumed to trickle down on their own.
Why China moved first
The numbers explain the urgency. China's core AI industry was worth more than 1.2 trillion yuan, about $178 billion, in 2025, with more than 6,200 AI companies. By the end of 2025, over 30 percent of manufacturing firms with annual revenue of at least 20 million yuan had adopted AI. That is a lot of deployment in sectors that employ a lot of people, which is why the policy talks about pace rather than bans.
It is also a different theory of adjustment than the one dominant in the US and UK, where the default has been to let adoption run and handle displacement through general labour policy. China is putting the assessment, the notice period and the retraining obligation on the employer, before the deployment rather than after the layoff.
What it does not do
It does not ban any use of AI, set a cap on deployment, or create an individual right to keep a job. The buffer period is something companies are "encouraged" to establish, not something they must. Enforcement runs through existing labour and market-access machinery, which varies by province.
It also does not answer the harder question for workers: who pays when a role disappears entirely rather than changing shape. Retraining into a different job is the plan's answer, and retraining has a mixed record everywhere it has been tried.
What to take from it
For anyone outside China, the useful read is the template. Employment-risk assessment before deployment, notice before the cut, retraining funded by the employer, and rules on algorithmic management. Those four items are the same asks unions and regulators are making in Europe, and China has now put them in a national document first.