In brief
- China’s government has told ministries and provinces to assess, monitor and plan for job losses caused by AI. The guidelines are policy guidance, not a law.
- The UK set up a unit in January to track AI’s effect on jobs, now part of a new AI Economics Institute. We found no published UK or Welsh plan for responding specifically to large-scale AI job losses.
- The only Welsh survey figures on AI at work we found are unweighted, and the Welsh Government says it needs better data. In our view, Wales should plan now for AI job losses, while it improves its data.
China has published a plan of work for protecting jobs as AI spreads. This explainer sets out what the document says, what it leaves open, what the UK and Wales track today, and what Wales could learn. Quotes from the Chinese text are in our own translation.
What China has issued
On 23 October, the General Office of China’s State Council published guidelines on adapting to artificial intelligence while supporting jobs. The State Council is China’s cabinet. The document is dated 16 October and numbered Guobanfa [2026] No. 29.
Officials are asked to weigh AI’s job risks, watch for them, and have a plan ready if they grow.
It is a 意见 (yijian), a form of policy guidance. Xinhua’s official English summary calls it “guidelines”. It is addressed to provincial governments and to the ministries and agencies of the State Council. It was issued “with the approval of the State Council”, in our translation. It is not a law.
Xinhua says it sets out “18 tasks in six areas”, and the Chinese text bears that out. The areas run from creating jobs with AI and training workers, through workers’ rights and social security, to monitoring and responding to job risks.
Officials are asked to weigh AI’s job risks, watch for them, and have a plan ready if they grow. Human resources departments and development and reform departments, at every level, are to co-ordinate the work. There is no official English version of the full text, so the translations in this piece are ours.
Five tasks in the government’s own words
Five of the 18 tasks carry the core. We quote each from the Chinese text, in our own translation.
Task 6 asks officials to “mitigate the job-displacing effects of AI applications”. AI should be rolled out first “in jobs where labour is scarce or conditions are hazardous”. In labour-intensive industries, officials should handle “the scope, intensity and pace of AI use” steadily and prudently, “so as to reduce the impact on workers as far as possible”.
Task 7 sets up “a mechanism for assessing the employment risks of AI applications”. For “AI application projects in market-access fields that may cause a relatively large shock to employment”, officials should “carry out an employment risk assessment at the same time as the project is reviewed”. Market-access fields are areas where a project needs official approval to go ahead. For other projects, assessments are to be carried out “as actual circumstances require”.
Task 12 covers help for people who lose work. Eligible workers displaced by AI should get “targeted careers guidance, skills training, job referrals, support for starting a business and other services”. Insured people who become unemployed should be paid unemployment insurance “promptly and in full”.
Task 16 is about monitoring. It calls for “cross-departmental sharing of data” and for “intelligent models for monitoring and analysing employment data by region and by sector”. The aim is to make “monitoring and early warning of employment risks more sensitive”.
Task 18 is the contingency plan. It brings AI job risks “within the scope of the mechanism for preventing and defusing the risk of large-scale unemployment”. It tells officials to “draw up risk response contingency plans in a co-ordinated way”. The Chinese word used, 预案, is the standard term for an emergency or contingency plan.
What the guidelines leave open
The best-known idea is voluntary. Task 6 says: “Encourage enterprises to explore setting up an unemployment buffer period system: when rolling out AI on a large scale, tell affected workers in advance and provide support for moving into new roles or new jobs.” It sets no length for the buffer. A buffer period is something firms are asked to consider, not something they must offer. China Newsweek reports that the idea appears in a national policy document for the first time.
On our reading of the full text, it sets no threshold, budget, deadline or penalty. It does not define a “relatively large” shock to employment. Caixin Global, a Chinese business publication, called it an “18-point plan to cushion workers from artificial intelligence-driven displacement”.
A buffer period is something firms are asked to consider, not something they must offer.
Work to put the guidelines into practice has begun. Hunan’s human resources department met on 2 November to plan the work, and said it would issue local implementing rules. On 3 November an official of the National Development and Reform Commission, China’s economic planning agency, said monitoring and early warning would be strengthened, Lianhe Zaobao reported. Zhejiang province has launched an AI training drive aiming at more than 2 million training sessions by the end of 2028, The Paper reported on 9 November. The report does not say the drive comes under the new guidelines. On 10 November the human resources ministry said it would “fully implement” them, in our translation.
In our searches of Chinese government websites and news reports up to 11 November, we found no national rule yet that defines a large shock, no published AI-specific contingency plan, and no count of assessments carried out. For now, on our reading, these are instructions to plan, not plans.
What the UK tracks
The UK does monitor AI’s effect on jobs. What it does not have, as far as we can find, is a plan like China’s.
In January 2026 the UK Government set up an AI and Future of Work Unit. Its job was to “research and monitor AI’s economic and labour market impacts to provide timely advice on when new policies should be implemented”, Civil Service World reported. The Department for Science, Innovation and Technology published an assessment at the same time. It said: “The available evidence does not yet provide clear answers to many of the questions that matter most for policy.”
In June the unit became part of a new AI Economics Institute, run jointly by HM Treasury and the science department. The institute will “strengthen the evidence base on how AI affects productivity, labour markets, and growth”. It says it “will publish selected analysis externally where it supports open debate”.
The available evidence does not yet provide clear answers to many of the questions that matter most for policy.
The January assessment listed six areas to track and said “specific metrics are under development”. It said the unit intends to publish updates on these indicators, subject to agreement across government. We could not find a published update of those indicators on GOV.UK by 11 November. The unit did publish a snapshot of entry-level hiring, with LinkedIn, in June.
The Office for National Statistics asks businesses about AI in its regular business survey. In September 2026, about 30% of UK businesses said they used at least one AI technology, according to the ONS. Among AI-using businesses with 10 or more employees, “6% reported reduced headcount as a result. Fewer than 1% reported that their headcount had increased.” Administrative and clerical roles were the ones most often named as affected.
The UK Government set up a unit in January 2026 to research and monitor AI’s effect on jobs, and in June folded it into a new AI Economics Institute. We found no published UK or Welsh plan, like China’s, for responding specifically if AI causes large-scale job losses, and no requirement to weigh job losses when approving AI projects.
What Wales knows
Wales has fewer numbers of its own, and the Welsh Government says so.
On 21 October the Welsh Government’s Cabinet Minister for Enterprise, Connectivity and Energy was asked in the Senedd about AI and jobs for young people. The minister said “the evidence is still emerging” and “we need to have a better understanding and better data.” The minister went on: “we need to get that kind of data here and we need to understand how this is playing out across different sectors and across different geographies. That work is ongoing”.
These are figures for businesses that responded, not for Welsh businesses as a whole.
The most recent Welsh figures on AI at work that we found come from the same ONS business survey, and they come with a warning. ONS publishes them unweighted: each responding business counts the same, whatever its size or sector. The Welsh figures cover businesses with a presence in Wales, including UK firms with a site there; 769 of them responded in late September and early October. Of those, 32.4% said they did not use AI, against 40.1% of respondents across the UK. Among those that used AI or were not sure, a smaller group, 7.5% reported reduced headcount, against 5.7% UK-wide.
These are figures for businesses that responded, not for Welsh businesses as a whole. The survey is voluntary and the Welsh sample is small, so the gaps with the UK may not hold up.
Older evidence on exposure to AI says little about Wales directly. A 2023 Department for Education study found that workers in London and the South East had the highest exposure to AI, and were “the only areas to be above the average for the UK as a whole”. Our inference is that Wales was below the UK average on that measure. The science department’s January assessment warned that exposure scores “have not been extensively validated against real-world outcomes”.
The AI Plan for Wales, published in November 2025, covers skills and adoption. We found no commitment in it to monitor AI’s effect on employment. A written question on how Welsh communities will benefit from UK investment in AI was answered on 30 October. The answer lists employment among the topics the Government is working on, but does not describe any monitoring of jobs.
The case against, and what Wales could learn
There is a serious case that planning for AI job losses now is premature, or will not work. It deserves a fair hearing.
First, there is little sign yet of a jobs shock. The Yale Budget Lab, which tracks the US labour market, said in September that its data “does not provide clear evidence of labor market disruption associated with AI.” In China, the economist Cai Fang of the Chinese Academy of Social Sciences has said that “so far there is no sign in the macro data of a clear employment shock from AI, but signs are emerging at the micro level”, in our translation of his words as reported by China Newsweek.
Second, AI’s effects are hard to pin down. The UK science department warned in January that “exposure is not adoption”. It said exposure is linked to other things, such as interest rates and business cycles, that could explain the same patterns.
Third, help for displaced workers does not always work. A US evaluation by Mathematica of Trade Adjustment Assistance, a scheme for workers who lost jobs to trade, found participants earned “about $3,300 less than their comparisons” in the final year studied. That study is from 2012.
In our view, AI job losses are coming, and Wales should plan for them now.
Finally, the Chinese text is soft. In our view, with no budget, deadline or penalty, how much it changes will depend on how ministries and provinces apply it. Xu Tianchen, a senior economist at the Economist Intelligence Unit, told Lianhe Zaobao, in a report published on 4 November: “It is not yet clear what new demand the AI era will create.” He also said job losses should be kept in check.
We take a different view. In our view, AI job losses are coming, and Wales should plan for them now. Cai Fang told a conference in Shanghai on 21 October that AI “is replacing jobs starting from relatively high-end posts”, in our translation, and he says signs are emerging at the micro level. The Welsh Government’s economy minister told the Senedd in October that AI is “one of the factors that is driving this unique level of disruption” in the labour market, while adding that the evidence is still emerging. Among UK firms with 10 or more staff that use AI, 6% said it had reduced their headcount and fewer than 1% said it had increased it, though most reported no change.
Cai Fang also says this brief quiet period gives “a rare chance to prepare our response in advance”, in our translation. In our view, the mixed record of help for displaced workers is a reason to design it well and test it, not to go without.
What could a Welsh plan contain? In our view, four things drawn from China’s tasks, without copying its methods.
The first is a contingency plan, agreed before it is needed. China folds AI job risks into a mechanism it already has for large-scale unemployment. Wales could set out now how its existing help for people who lose their jobs would scale up if AI losses grew quickly, and who would lead.
The second is early notice. China’s buffer period is voluntary, but the idea behind it is sound: warn people before AI changes their jobs, and help them move. The minister said the Welsh Government is working with employers, trade unions and education providers on how AI is changing recruitment, training and workplace learning. They could agree what fair notice looks like when an employer adopts AI at scale.
The third is retraining while people are still in work. Zhejiang plans more than 1 million in-work training sessions over three years in industries “vulnerable to AI shocks”, in our translation. Wales could aim the personal learning accounts and AI skills pilot the minister mentioned first at administrative and clerical roles, the ones most often named as affected in the ONS survey.
The fourth is numbers to steer by, so Wales knows when to act. That means weighted Welsh figures on AI use and headcount, the AI jobs indicators the UK Government said in January it intended to publish, and a date for the Welsh Government to report back to the Senedd on its data work.
That is a case for policy, not for any party. In our view, the time to prepare is while the figures are still quiet.
Words used here
- 意见 (yijian)
- a Chinese policy document giving official guidance. Xinhua translates it as “guidelines”. It is not a law.
- State Council
- China’s cabinet. Its General Office issued these guidelines.
- 预案 (contingency plan)
- a plan drawn up in advance for how to respond if a risk becomes real.
- Market-access fields
- areas where a project needs official approval to enter the market.
- Unemployment buffer period
- an arrangement in which a firm tells workers in advance that AI will affect their jobs, and helps them move into new roles. Voluntary in these guidelines.
- Unweighted figures
- survey results in which every responding business counts the same, whatever its size. They describe the businesses that answered, not all businesses.
- AI exposure
- how much of a job’s tasks AI could in theory do. It is not a forecast of job losses.