The Chaoyang District People’s Court in Beijing has recently decided a labour dispute arising from the use of AI tools in the course of work. An employee of an insurer, Ms Wu, used an AI tool to assist her work. The company inferred from the general rule that the AI platform retains data that she presented a risk of disclosure, found that this constituted serious breach of its internal rules, and unilaterally terminated the employment contract. The court did not uphold the dismissal: at first instance it found the termination unlawful and ordered that the notice of termination be set aside, that the employment contract continue to be performed, and that the shortfall in the year-end bonus, performance pay and lost wages be made good. The judgment has taken effect.
The case is not complicated in itself, but it places the new fact of “an employee using an AI tool” squarely within the existing tests of labour law. The court’s approach is of direct value to companies now drafting AI usage policies.
Facts and issue
The essential facts are these: the employee used an external AI tool to assist her work; the company subsequently concluded that the conduct carried a risk of disclosure and dismissed her on that basis.
The issue turned on a single question: did the use of an AI tool constitute “serious breach of the employer’s rules and regulations” within the meaning of Article 39(2) of the Labour Contract Law? That provision permits dismissal without statutory compensation where the employee seriously breaches the rules, and the threshold is accordingly set high.
The company’s case rested on an inference: because AI platforms generally retain data, use of the platform necessarily entails a risk of disclosure. The difficulty is that the inference moves directly from a general attribute of the platform to a specific risk on the part of this employee, omitting the essential intermediate step — what the employee actually did.
Why the court rejected the inference
The court’s conclusion may be summarised in one sentence: the mere existence of a general rule that AI platforms retain data, and an inference of disclosure risk drawn from it, are insufficient to establish that the employee’s conduct amounted to serious breach of the rules.
That conclusion has two limbs.
The first concerns the object of proof. The proposition that “AI platforms generally retain data” is a statement about the attributes of a tool, not about the conduct of the employee. To establish a disciplinary breach, what must be proved is the employee’s specific conduct: what information was entered, whether that information fell within the company’s confidential material, and whether actual or probable harm resulted. The material on which the company relied in support of the dismissal omitted precisely this.
The second concerns the method of inference. In a labour dispute the employer bears the burden of proving that a dismissal was lawful. To reason backwards from the general risk of a tool to the individual fact of misconduct is, in substance, to substitute inference for evidence. Such an inference is particularly untenable where a confidentiality obligation is in issue: a breach of that obligation must be tied to specific information, not to the nature of the tool.
The review standard established
The practical value of the judgment lies in the points of review it signals for cases of this kind.
Review begins with the basis in the rules. The employer must show that valid rules exist, adopted through a democratic procedure and published to employees, and that those rules contain an express prohibition and a specified consequence in respect of the conduct in question. Where the rules state only, in general terms, that “company information must not be disclosed”, without making any provision for the use of external AI tools, then “use of AI is a disciplinary breach” lacks any normative foundation.
The second point is the correspondence between conduct and consequence. Even where a relevant rule exists, it must still be shown that the conduct reached the threshold of “serious” and that it corresponds to a specific clause. To classify every question concerning AI use as “serious” will rarely be upheld in an individual case.
The third is the actual content of the evidence. On the facts of this case, an employer asserting that the use of AI created a risk of disclosure would need at the least to fix the record of the employee’s specific operations, to establish the confidential character of the information concerned, and to explain the harm or the real risk of it. Such material is the foundation of the decision to dismiss and the preparation for any subsequent arbitration or litigation.
As to the consequences of an unlawful dismissal, Articles 48 and 87 of the Labour Contract Law provide that the worker may require continued performance of the contract, and the employer shall continue performance; where continued performance is not sought or is no longer possible, the employer shall pay compensation at twice the statutory rate. In this case the employee elected for continued performance, and that election was upheld.
A parallel line: replacement of roles by AI
Alongside this case lies a more common category of dispute — dismissal on the ground that the role has been replaced by AI. Here too the employer’s position is weak.
In a Beijing arbitration, a technology company that had moved wholly to automated AI data collection abolished the employee’s department and role and terminated the contract on the ground that “the objective circumstances at the time of contracting had materially changed”. The arbitration commission found the termination unlawful, on the basis that the arrangement in substance transferred the risk of technological change to the worker. In a Hangzhou case, the finding of unlawful termination was upheld on appeal in April 2026, and the company was ordered to pay more than RMB 260,000. The Guangzhou Intermediate People’s Court took the same position in a 2024 case, holding that the adoption of AI was an adjustment made by the enterprise within its autonomy in response to market change and did not fall within “objective circumstances”.
Taken together, the two categories point to a single boundary: technological change is not in itself a proper ground for dismissal, and cannot replace proof of the specific facts. Whether the question is “an employee breached the rules by using AI” or “the role has been replaced by AI”, the employer must return to the facts and to its own rules, and may not rely on a general assessment of a new technology.
Implications for workforce management
The most immediate implication is that a company needs an AI usage policy that can actually be used. “Usable” means answering at least the following: which tools may be used and in which scenarios they are prohibited; which data may be entered into external AI and which may not; whether high-risk use requires approval and how usage records are kept; and how the consequences of breach are graded. Without such content, a policy can scarcely be invoked in a dispute.
The second is the building of evidentiary capacity. The company in this case failed largely because of the structure of its evidence: it proved the attributes of the platform but not the conduct of the employee. Operation logs, approval trails and information classification are necessary both for day-to-day management and as preparation for dispute.
The third is advance arrangements for confidential roles. For positions with genuine access to core secrets, the safer course is to control risk at the front end through dedicated confidentiality agreements, whitelisting of tools and devices, and data loss prevention measures, rather than to dismiss after the event on the ground that “the platform is risky”.
For assistance with the drafting and compliance review of an AI usage policy, the revision of an employee handbook, or the conduct of a labour dispute, see the employment compliance service page. Related reading: the commentary on the Supreme People’s Court opinion on AI disputes and the analysis of the Action Plan for AI Plus Software discuss, respectively, liability in AI scenarios and the policy direction for employment.
This article is prepared on the basis of public reports of the judgment of the Chaoyang District People’s Court in Beijing (October 2026), the relevant Beijing arbitration and the Labour Contract Law of the People’s Republic of China. It is for reference only and does not constitute legal advice. Individual matters should be assessed on their own facts and evidence.