--- title: "China Added a Job-Loss Check to AI Project Approvals" date: 2026-10-28 category: China AI Decoded site: NeuroAI canonical: https://neuroai.site/a/na-decode-ai-employment-check language: en --- # China Added a Job-Loss Check to AI Project Approvals > A State Council General Office opinion puts a labour test inside AI project review and tells agencies to weigh the pace of adoption in labour-intensive industries. The opinion was posted on a Friday in late October, and the clause that matters sits in the middle, under a heading about upgrading. It does something unusual for a document of its kind: it asks officials to slow down. Not everywhere, and not permanently. But in the industries that employ the most people, the pace of automation has become something a department is told to weigh before it signs. For several years the working assumption was that adoption was the goal, and that displacement was a transition cost to be settled afterwards. The October opinion reverses that order, at least on paper. ## Key takeaways - China's State Council General Office issued 国办发〔2026〕29号 on 16 October 2026 and published it on 23 October, with 18 tasks in six areas (State Council, gov.cn). - Item 7 requires an employment-risk assessment at review stage for market-access AI projects likely to cause a large employment shock (State Council General Office, 16 October 2026). - Item 6 tells agencies to control "the scope, intensity and pace" of AI in labour-intensive industries and to minimise the impact on workers. - Item 18 places AI displacement inside China's existing mass-unemployment risk mechanism, which predates the AI debate. - The People's Bank of China wrote on 19 October 2026 that AI had not yet shocked total employment but had begun to change its structure. ## The clause that changes the sequence The State Council General Office (国务院办公厅) signed the opinion on 16 October 2026 and released the text on 23 October (gov.cn). Xinhua counted 18 numbered tasks grouped into six areas: creating jobs, building an employment-friendly upgrade path, training, public employment services, worker protection, and monitoring (Xinhua, 23 October 2026). The sequencing is the story. Item 7 — 建立人工智能应用就业风险评估机制, the employment-risk assessment mechanism — puts a labour test inside project review rather than after it. For AI application-scenario projects in market-access (市场准入) fields that could cause a relatively large employment shock, the assessment runs at the same time as the review, and the relevant departments are to coordinate retraining and redeployment support (State Council General Office, 16 October 2026). That is a different instrument from a subsidy. A subsidy shapes what gets built; an assessment shapes what gets cleared. For a local government that has spent years competing for AI projects, it also creates a document that someone has to write and sign. Item 6 supplies the throttle. Agencies are to "prudently grasp the scope, intensity and pace" of AI application in labour-intensive industries and to "minimise the impact on workers" (State Council General Office, 16 October 2026). The same item encourages firms to explore an unemployment buffer period (失业缓冲期) — advance notice to affected workers, plus retraining and redeployment support — but the verb is "encourage", not "require". No penalty is specified for a firm that declines. ## Where the state still pushes Nothing in the document slows the money. Item 3 names two government funds as instruments for AI employment — the National Venture Capital Guidance Fund and the National AI Industry Investment Fund (国家人工智能产业投资基金) — and asks them to invest on market terms. Item 1 divides the map: eastern provinces are encouraged to site core industry, while central and western ones are guided to take on data annotation (数据标注) and device manufacturing. So the same document carries a brake and an accelerator, aimed at different places. Three consequences follow. - Approvals slow where headcount is high and speed up where labour is scarce. Item 6 also says AI should be deployed first in roles that are short of workers or physically dangerous — a priority lane for deployments that attract no objection. - Data annotation is now framed as an employment programme, not only an industrial one. That raises the political cost of automating annotation work itself, which supplies much of the training data behind robotics. - The assessment becomes an audit trail. Once a project carries a signed employment assessment, a later closure has a document to be measured against. ## The number nobody has published The policy is anticipatory, and the government says so. At the Financial Street Forum in Beijing, in a speech text the central bank published on 19 October 2026, the People's Bank of China wrote that AI "has not yet caused an obvious shock to total employment, but has already begun to affect employment structure" (中国人民银行, 19 October 2026). That sentence matters twice over. It concedes structural change while declining to quantify it, and it comes from the institution that runs the payment and credit-reporting systems. No national displacement figure appears in the opinion, in the Xinhua readout, or in the central bank's remarks. What exists instead is machinery: item 17 orders a dedicated survey of AI's employment impact, standing tracking of penetration and substitution, and a set of AI employment research centres attached to universities (State Council General Office, 2026). The first survey will be the first number, and it has not been fielded yet. Until then, the only honest reading is that China has built a process for deciding whether automation is moving too fast, and has not yet decided what "too fast" means. ## Honest limitations Everything above about effect is inference from text, not observation; the opinion had been public for five days when this was written, and no province has published an implementing plan. The document's force is uneven — "encourage", "guide" and "explore" carry most labour provisions, while only the assessment and monitoring duties read as obligations. The central bank's line is a judgement, not a measurement, and it addresses total employment rather than any sector. Whether an employment assessment can stop a project, or merely annotate it, is not stated anywhere in the text. ## What readers can do now - Ask a Chinese partner whether its next deployment falls in a market-access field; that classification, not the technology, decides whether an employment assessment is triggered. - Watch provincial human-resources and development-reform departments for the first implementing notices; their wording will show whether the assessment is a gate or a formality. - Treat the item-17 survey as the base case for any China displacement estimate through 2027, and discount consultancy figures until it appears. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-decode-ai-employment-check Free to quote with attribution and a link to the original.