A Firing That Became a Test Case
The case that crystallized the problem was mundane on its face. Zhou, a 35-year-old quality-control supervisor who reviewed AI-generated customer responses at a Hangzhou fintech firm, was offered a demotion and a pay cut from 25,000 yuan to 15,000 yuan a month, with the company arguing AI had made his old role redundant. He refused. The company fired him.
The Hangzhou Intermediate People's Court didn't accept that reasoning. In a ruling reported by the South China Morning Post on May 3, 2026, the court held that swapping a worker for AI on cost grounds does not meet the Labor Contract Law's threshold of a "material change in objective circumstances" required to justify termination, and ordered the firm to pay Zhou more than 260,000 yuan (~$38,000) in compensation. A judge on the panel added a line that will outlast the case itself: "We don't believe AI technology has reached the point where it can substantially replace human workers."
That is a striking thing for a court to assert as law rather than fact — and, per an SCMP report on June 14, 2026, it's one node in a fast-growing docket of AI disputes that Chinese courts are resolving one at a time, with no AI-specific statute to guide them.
Judges Are Filling a Statutory Vacuum
Because China has no comprehensive AI liability law, judges are defaulting to Article 1165 of the Civil Code — the general fault-based tort standard — and building doctrine case by case. In January 2026, the Hangzhou Internet Court dismissed a suit against an LLM provider whose chatbot had hallucinated a wrong campus address and then "promised" a 100,000-yuan payout if it was mistaken; the court held that a generative AI service isn't a "product" under the Product Quality Law, so strict liability doesn't apply, and that the hallucinated promise didn't bind the provider's actual intent. Different courts, different questions: the Guangzhou Internet Court has separately heard roughly 700 face-theft suits over three years, as a booming market for licensing people's likenesses to AI-generated microdramas collides with biometric-privacy claims. Each ruling is defensible on its own facts. None of them is binding precedent on the next court to hear a similar dispute, because China's civil-law system gives "guiding cases" persuasive weight at best.
The Case for Letting Courts Lead
There's a real argument for exactly this approach. Statutes calibrated too early ossify around today's technology and struggle to bend for tomorrow's; a body of case law lets rules emerge from real disputes rather than legislative guesswork about hypothetical harms. China's own State Council seems to appreciate that tradeoff: its 2026 legislative work plan, published May 11, commits only to "accelerating comprehensive legislation" for AI's "common elements" — data, computing power, algorithms, IP, cybersecurity, supply-chain security — without locking in liability rules, while the Supreme People's Court has separately said it will spend the 2026–2030 plan period developing judicial guidance on data rights and AI-generated content rather than rushing a code. Even Brussels, no stranger to precautionary regulation, abandoned its own AI Liability Directive in February 2025 after concluding, in the Commission's words, there was "no foreseeable agreement" — a tacit admission that legislating AI civil liability well is genuinely hard, and getting it wrong can lock in worse rules than letting doctrine develop first.
Why the Gap Still Needs Closing
But there's a difference between deliberate calibration and a vacuum, and China is closer to the latter. A company deciding whether it can restructure headcount around an AI tool cannot look up a settled duty-of-care standard; it has to guess which of several unweighted tort theories a given intermediate court might apply, in a system where similar facts can plausibly draw different outcomes in Hangzhou, Guangzhou, or Beijing. That unpredictability is itself a tax on the investment China says it wants: deployment decisions get more conservative, and disputes that a clear default rule could resolve quickly instead consume years working through arbitration, trial, and appeal, as the Zhou case did. A unified statute doesn't have to be maximalist to fix this — it mainly needs to settle the standard-of-care and burden-of-proof questions courts are currently answering ad hoc, leaving room for judicial guidance to handle the specifics. That is closer to what Beijing has actually signaled than a full EU-style liability code, and it's the right target: enough certainty for firms to plan around, without freezing rules around 2026's technology.
"AI replacement, notably, does not automatically justify terminating a labor contract," Zhejiang Xingjing law firm's Wang Xuyang told Xinhua after the ruling — the kind of settled-sounding principle that, absent a statute, only exists because one intermediate court happened to say so first.