A Campaign, Not a Court Case
On September 2, 2026, the Cyberspace Administration of China (CAC) reported the results of the second phase of its "Qinglang: Rectify AI Application Chaos" special action: 5.61 million pieces of "illegal and non-compliant" AI-generated content removed, more than 49,000 accounts penalized, and over 2,400 websites and apps sanctioned since the campaign launched in April 2026. Major platforms — Douyin, Kuaishou, Weibo, WeChat, Baidu, Bilibili, RedNote — and AI assistants including Doubao, Yuanbao and Qwen were named as having tightened filters in response (Xinhua; SCMP).
The legal hooks are real statutes, not improvised authority. The 2023 Generative AI Measures (effective August 15, 2023) require providers with "public opinion" or "social mobilization" capacity to register their models and label synthetic output. The AI-Generated Synthetic Content Labeling Measures, jointly issued by the CAC, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the National Radio and Television Administration, took effect September 1, 2025, mandating explicit and embedded metadata tags on AI-generated text, images, audio and video (CAC notice). This crackdown is, on paper, enforcement of a year-old labeling mandate — not a novel censorship instrument invented for the occasion.
The Case Regulators Would Make
Give the CAC its due: the underlying problem is real. Chinese platforms have been flooded with what regulators call "digital slop" — AI "remixes" of classics like Journey to the West twisted into clickbait, synthetic disaster footage designed to panic users, voice-cloned scams targeting the elderly, and face-swapped videos impersonating real people without consent. Every jurisdiction now regulating synthetic media — the EU's AI Act transparency obligations for deepfakes, several U.S. states' election-deepfake statutes — starts from the same premise: unlabeled synthetic content that looks real causes concrete harm, from fraud to reputational damage to public panic. A rule requiring a platform to tag an AI-generated video as AI-generated, and to take down synthetic content used for fraud or non-consensual impersonation of minors, is not an unreasonable ask of a AI industry moving this fast. Proportionate, narrowly targeted enforcement against fraud, non-consensual deepfakes, and undisclosed synthetic media protecting consumers is a legitimate regulatory function, and one this publication has argued other jurisdictions should pursue more consistently.
Where the System Breaks Down
The problem is not the labeling requirement itself — it's what sits underneath the enforcement. The seven violation categories the CAC cites mix genuinely narrow harms (child sexual content, financial fraud) with categories elastic enough to swallow ordinary speech: "fabricating and spreading false information," "network astroturfing," and — the vaguest of all — content contributing to "digital slop." Who decides that an AI-generated retelling of a classic novel is degraded content rather than legitimate transformative work? The CAC does, unilaterally, with no published adjudication standard, no case-by-case reasoning released for the 49,000 penalized accounts, and no independent tribunal a platform or creator can appeal to. Compare that to the EU's AI Act, where transparency obligations for synthetic content are enforceable through national regulators subject to judicial review in ordinary courts, or to the handful of U.S. state deepfake statutes, which create private rights of action litigated in front of judges who did not write the rule. China's enforcement apparatus is investigator, prosecutor, and judge in one body, moving at a scale — 2,400-plus platforms disciplined in four months — that leaves essentially no room for a wrongly-flagged account to contest the call.
That asymmetry matters more, not less, because the underlying rule is legitimate. A vague standard enforced by an unaccountable actor is dangerous precisely because it can't be distinguished, from the outside, from a narrow standard enforced fairly — until the day it's used against a journalist's AI-assisted investigation, a satirist's deepfake parody of an official, or a foreign platform's compliance posture becomes politically inconvenient. The same infrastructure built to catch romance-scam voice clones is equally capable of throttling any AI tool that produces content Beijing doesn't like, with the same 561-figure removal count reported either way.
Why This Travels Beyond China
The stakes extend past China's borders. As governments everywhere — including democracies — look for templates to regulate synthetic media, China's model offers a tempting shortcut: broad executive discretion, platform-level pre-filtering obligations, and enforcement theater backed by big numbers, all without the friction of judicial review. The lesson for regulators watching Beijing's approach isn't that content labeling for AI is wrong — it's that labeling mandates only remain proportionate if the enforcement mechanism includes independent review, published reasoning, and a route to appeal. Strip those out and even a defensible rule against deepfake fraud becomes a general-purpose speech control with a compliance rate the CAC gets to report on its own terms.