Court orders platform to reverse an AI-label penalty it failed to explain

A platform hid a short answer and muted its author for a day after classifying the text as unlabeled AI-generated content. The author, identified as Tang, maintained that he had written it himself. An unsuccessful appeal led to a lawsuit, and the court ordered the platform to restore access to the answer and remove the associated violation record. Beijing Internet Court presented the case on November 10. As disputes over AI-edited commercial images and synthetic video continue, the case raises the reverse problem: what happens when a person disputes being classified as a machine?

The platform required creators to label AI-generated material. It defended its decision by saying the answer had passed through both automated identification and human review. But it also argued that technical secrets and algorithmic logic meant it did not need to explain the mechanism to the user. The practical stakes extended beyond a single missing post. Tang had lost the ability to speak for a day and had a violation recorded against the account.

The court accepted that a platform may review and handle AI-generated material under its contract with users. It nevertheless required a reasonable basis for the decision. According to Beijing News reporter Zhang Jingshu’s account, the short answer was a form of spontaneous writing, making a demand for drafts or source files unrealistic. The platform controlled the algorithm, its operation and the review results. It therefore had to provide reasonable evidence or an adequate explanation. Its failure to do so made the unsupported restriction a breach of contract.

Think of the classifier as a detector that automatically issues a ticket. An alarm still leaves questions about what it measured, why that supports a penalty and how the recipient can challenge it. Protecting the detector’s design is different from explaining a particular ticket. The case addressed evidence and explanation in a platform decision, rather than establishing a particular AI detector’s accuracy. The court said users should be able to understand the explanation and the route for correcting a mistaken classification.

For creators, retaining the appeal history and the platform’s stated reasons may be more useful than repeatedly debating whether a paragraph sounds robotic. For platforms, human review needs to amount to more than another stamp on an automated result. At the same briefing, the court recommended giving reasons for penalties, offering accessible appeals and matching restrictions to the seriousness of the violation. Identifying generated material is only one part of moderation; people on the receiving end also need a workable way to question and correct the decision.

Original source published: The Paper, November 10, 2026, 19:48 Beijing time / November 10, 2026, 6:48 AM EST