An employee whose job was to check the work of an artificial intelligence was dismissed when that artificial intelligence became able to do it alone. He sued his employer. He won.
The case comes from Hangzhou, and it is circulating widely under a misleading formula: “China bans AI from taking human jobs”. No ban has been passed. But what the Chinese courts have established is more interesting than a ban — and it puts France, on this precise point, in exactly the opposite position.
The Zhou case, in detail
The facts are documented by the Chinese business press and picked up by Bloomberg, NPR and Fortune.
Zhou is hired in November 2022 as a quality control supervisor at a technology company in Hangzhou. His salary: 25,000 yuan a month. His job: to check the relevance and compliance of the answers produced by the company’s large language models. In other words, an occupation born entirely of AI, and whose very existence depends on AI’s imperfections.
In 2024, the company decides its own systems can now carry out that checking. It offers Zhou a lower-level operational post paying 15,000 yuan — a cut of 40%. He refuses. The company then unilaterally terminates the contract, citing a restructuring and a headcount reduction linked to the introduction of AI.
Zhou goes to the labour arbitration commission, which finds in his favour. The company appeals. The Hangzhou intermediate people’s court confirms: the dismissal is unlawful. The company is ordered to pay more than 260,000 yuan, around 33,700 euros at the August 2026 rate.
The ruling is published on 30 April 2026 — the eve of 1 May — in a collection titled “typical examples of protecting the rights of companies and workers in the AI era”. The choice of date is probably not accidental.
The reasoning: imposed circumstance or voluntary decision?
This is the heart of the case, and it is what makes it something other than a judicial news item.
Chinese employment law allows an employer to terminate a contract in the event of a major change in objective circumstances. This is article 40 of the labour contract law of the People’s Republic of China, which requires thirty days’ written notice or a month’s pay, and above all two cumulative conditions: the change must make the contract impossible to perform, and consultations on amending the contract must have failed.
The notion covers events imposed on the company. The cases accepted in practice are limited: force majeure linked to a natural disaster (earthquake, fire, flood); a change of law, regulation or public policy causing relocation, asset transfer, cessation of production or change of ownership; or a change in a franchisee’s scope of business. None of those looks like a decision to automate.
The question put to the court was therefore simple: does the arrival of AI constitute such a change?
The answer: no. The judges held that adopting an artificial intelligence system is a conscious and voluntary decision by the company, a strategic choice — not an adverse circumstance it suffers. A company therefore cannot invoke its own decision to automate in order to justify removing a post automatically.
The legal test adopted deserves quoting precisely, because it is the one that will set the precedent: to come within article 40, a change must be irreversible and unforeseeable. Technological change is neither — it falls, according to the Chinese courts that have ruled, within a company’s normal operating risks. A legal analysis published in February 2026 had already summed the position up in a phrase: adopting AI is a normal commercial decision and a proactive innovation, not an unforeseeable objective circumstance.
The court added a second ground, just as important: a redeployment offer carrying a 40% pay cut is not a reasonable offer.
In legal terms, that amounts to placing the risk of automation on the company rather than on the worker. It is a choice about who bears the risk, and it is explicit.
The paradox of Zhou’s job
One detail of this case is worth pausing on, because it heralds a whole category of disputes to come.
Zhou’s job did not exist before AI. Quality control supervisor for the output of a language model: that post was created precisely because models get things wrong, hallucinate, produce non-compliant answers. It belongs to that layer of jobs born of the need to correct the machine — annotators, evaluators, moderators, output reviewers.
Those jobs have a singular property: their existence is indexed to the flaws of the system they serve. The better the model gets, the less it needs checking. And the improvement comes in part from the checking work itself. In other words, these workers produce, through their daily work, the data that makes their own post redundant.
That is not an irony of circumstance, it is a structure. It explains why these posts will be hit first, and why disputes of Zhou’s kind will multiply before reaching other occupations. We noted the same mechanism when analysing why layoffs attributed to AI often rest on a declared reason rather than a verified cause: automation rarely advances where it is announced, and very fast where nobody is looking.
No, China has not “banned” anything
The formula going around has to be corrected, because it distorts the whole file.
No law has been passed. No general ban on replacing workers with machines exists in China — and Chinese companies are automating massively, as everywhere else. These are court rulings handed down in individual disputes.
The next nuance matters, and it is specific to the Chinese system. The Hangzhou ruling was published as a “typical example”. That is not the same thing as a “guiding case” (指导性案例), a category selected by the Supreme People’s Court which courts across the country must follow where a case presents similar facts. A typical example published by an intermediate court has strong guidance value, but not the binding reach of a guiding case.
In concrete terms: a Chinese company can still automate a post. What it can no longer do without risk is use that automation as an automatic ground for dismissal in order to get around the protections of the employment contract.
That is narrower than a ban. It is also far more operational.
Three cities, three converging rulings
What gives the case its weight is that it is not isolated.
- Hangzhou: the Zhou case, quality control supervisor, ruling of the intermediate court.
- Guangzhou: the intermediate court upholds the first-instance decision finding unlawful the dismissal of a graphic designer whose work had been taken over by AI.
- Beijing: in the case known as “Liu”, an employee dismissed after the automation of map data collection also wins, after passing through the arbitration commission and then two levels of court. That case is earlier than the Hangzhou one: it was already being discussed by Chinese employment law practitioners in February 2026. Hangzhou did not open the way — it made it visible, by publishing its ruling as an example intended to guide other courts.
Three courts, three major cities, one line: automation cannot be used to get around employment law protections. It is that convergence, more than any single ruling, that constitutes the signal sent to employers.
That signal comes with a manual. Legal analyses written for companies operating in China now recommend four routes to automating without litigation: negotiate an amendment to the contract, offer training to the worker concerned, arrange an internal transfer, or conclude a settlement agreement. In other words, what the courts are closing is not automation: it is the shortcut.
In France, the law says exactly the opposite
This is the part that concerns the European reader directly, and the contrast is striking.
Article L1233-3 of the French labour code lists the grounds that can support an economic dismissal. There are four: economic difficulties, technological change, reorganisation necessary to safeguard competitiveness, and cessation of business.
Technological change is therefore in the law in black and white. And the case law is long-standing: the Court of Cassation has accepted since the 1990s that computerising an activity constitutes technological change capable of justifying an economic dismissal. The introduction of an AI fits into that category without difficulty — the lawyers who have written on the subject are unanimous.
In other words: where the Chinese court holds that adopting AI is a voluntary choice which does not justify a dismissal, French law makes it an explicit lawful ground.
That does not mean the French employer has a free hand. Three safeguards apply:
- It must establish the reality of the technological change — a vague project is not enough.
- It must show its direct effect on the post removed.
- It must have met its obligations to adapt and redeploy, which includes training the worker on the new tools.
Case law has clarified a point that now takes on its full meaning: technological change must be distinguished from a simple management choice, and the employer must justify why it makes removing the post necessary, not merely that it makes it possible. That is exactly the question the Hangzhou court decided the other way.
You can see both the proximity and the gap. The two systems are asking the same thing — is this a constraint or a choice? But they draw opposite conclusions: in France a poorly evidenced management choice weakens the dismissal; in China the fact that it is a choice is enough to invalidate it.
One more thing no source allows us to assert today: no Court of Cassation decision deals specifically with a dismissal motivated by the introduction of a generative AI. The reasoning rests on the analogy with computerisation. The first published case will be worth watching closely.
And Italy decided the other way
A European point of comparison exists, and it is almost contemporaneous.
In November 2025, a court in Rome ruled on the dismissal of a graphic designer — the same occupation as in the Guangzhou case. Her residual tasks had been handed to the marketing team, which “also uses the support of artificial intelligence”.
The court upheld the dismissal, under the justified objective ground provided for in Italian law, having checked that the employer had indeed looked for redeployment. In that reasoning, AI appears as a factual element of context in the restructuring, and not as an autonomous legal cause.
Two almost simultaneous cases, two identical occupations, two opposite outcomes. The difference lies not in the facts but in the question each judge is willing to ask: Rome examines the procedure the employer followed, Hangzhou questions the nature of the decision to automate.
Why now, and what it says about the Chinese context
These rulings do not fall into a vacuum. They arrive in a labour market under strain.
Unemployment among 16-24 year olds excluding students reached 16.9% in March 2026, against 16.1% the month before. 12.22 million graduates entered the labour market this year, 430,000 more than in 2024. Add trade tensions with the United States, deflationary pressure, and an accelerating technological transition that is slowing hiring.
In that context, protecting salaried employment becomes a question of social stability. Which casts a particular light on the Chinese industrial strategy we analysed in detail: the country gives its best AI models away free to the whole world in order to capture the global ecosystem, while judicially slowing the effects of that same technology on its domestic labour market. Accelerate for export, cushion at home.
The country is also running an active international AI governance policy, of which we documented a striking episode with the creation of a world organisation in Shanghai.
The limit these rulings do not cross
Here is the point the enthusiastic coverage systematically leaves out, and it is decisive.
These judgments rest on the employment contract. They protect employees. Yet China had in 2024 more than 200 million so-called “flexible” workers: delivery riders, gig providers, content creators, freelancers, subcontractors. These people have no employment contract in the classic sense.
They are, however, the ones most exposed to automation, and the ones with the least bargaining power. The case law of Hangzhou, Guangzhou and Beijing does not cover them.
That is exactly the mechanism we found when analysing what studies on jobs at risk actually measure: the protection goes to those already inside, while the adjustment falls on the most precarious and on those who never get in. A court ruling favourable to a qualified supervisor on 25,000 yuan a month changes nothing for a rider paid by the drop.
If this concerns you in France: what can be checked
The legal contrast is interesting, but the practical question matters more. Here is what French law makes it possible to check, without extrapolation.
The reality of the change must be established, not asserted. An employer cannot simply announce that it is “deploying AI”. It has to show an effective technological transformation. A software licence bought and barely used, an abandoned pilot project, a tool deployed after the dismissal was notified: all of these weaken the ground.
The link with your specific post must be shown. The change must have a direct effect on the post removed. That the company automated some other function is not enough.
The obligation to adapt and redeploy is a major checkpoint. The employer must have sought to adapt you to the new tool and to redeploy you to an available post. That is often where procedures fail: training a worker to work with AI costs less than a contested dismissal, and the absence of any attempt at training is noticed.
A large pay cut is not a neutral offer. That is the second ground adopted in Hangzhou. Under French law, pay is an essential element of the contract: a substantial change cannot be imposed, and the worker’s refusal is not in itself misconduct.
Finally, the distinction between technological change and management choice is your best angle. The employer must explain why the change makes it necessary to remove the post, and not merely that it makes it possible. That is precisely the line the Hangzhou court relied on to decide the other way.
One last element, often ignored: the European AI regulation imposes obligations when a system is used for employment decisions, those uses being classified as high-risk. That does not concern the ground for dismissal, but rather the tools used to assess, select or sort workers — a point whose timetable we set out in our analysis of what changes with the AI Act.
What to take away
The fact is real and verified. Three Chinese courts — Hangzhou, Guangzhou, Beijing — have found unlawful dismissals motivated by the replacement of the worker with an AI. In the best-documented case, the company was ordered to pay more than 260,000 yuan, around 33,700 euros.
But the formula going around is false. China has banned nothing. There is no law, no general prohibition. There are court rulings, one of them published as a typical example — strong guidance, not binding norm. What is prohibited is not automation, it is its use as an automatic ground for dismissal.
And the comparison with France is counter-intuitive. On this precise point, French law is less protective: technological change is explicitly among the grounds for economic dismissal in article L1233-3, and case law has accepted it for thirty years. The French worker is not defenceless — the reality of the change, its effect on the post and the redeployment obligations remain under judicial control — but the principle itself is reversed.
That leaves the question the Chinese text poses and that nobody in Europe has yet decided: when a company freely decides to automate, who should bear the cost of that decision? Hangzhou answered the company. Rome answered the worker. France, for now, has not been asked the question in those terms. It will be — and the answer will weigh well beyond one employment tribunal case, in a country where the trajectory of AI on employment remains largely unwritten.
The figure to keep
200 million. That is the number of “flexible” workers counted in China in 2024 — riders, freelancers, gig providers — whom these court rulings do not protect, for want of an employment contract. They are also the first exposed to automation. The most worker-friendly case law in the world is worth only what it is worth to those who are employees.
To follow the real capabilities of the models driving these transformations, our AI ranking is updated continuously.