Beijing Targets Fake Personas Hiding Behind AI Disclaimers
China's internet regulator has told platforms that an "AI-generated" or "fictional" tag is no excuse for hosting fake personas. Other labelling regimes, from Brussels to Brasília, face the same question: what does a disclosure actually cure?
The Beijing apm shopping centre on Wangfujing Street, Beijing, 26 December 2021 (illustrative).
Photo: N509FZ, CC BY-SA 4.0 · source
China’s top internet regulator has opened a two-month national campaign against fake personas on short-video platforms, and one of its targets is the very label Beijing made compulsory a year ago. In a notice published on 30 September 2026, the Office of the Central Cyberspace Affairs Commission (the CAC) said platforms may not rely on tags already applied to a video, such as “fictional performance”, “AI-generated”, “marketing” or “personal opinion”, or on boilerplate disclaimers, to sidestep content governance (in Chinese; translation by AI & GPP). For governments that have just switched on disclosure rules, the lesson is awkward: a label can say what something is without stopping the harm.
What the “Qinglang” Campaign Goes After
The campaign, branded “Qinglang · rectifying fake personas in short video”, lists five problem areas. Most involve staged human performance, from struggling delivery riders and farmers to fake doctors, rented luxury and staged rescues.
The third category is about AI. It covers AI-synthesised audio and video impersonating public figures for false endorsements, accounts posing as people in trending news, and account names hinting at a celebrity’s authorisation. A separate measure singles out AI-built “domineering CEO” and “health master” personas used to lure elderly viewers.
The CAC will punish accounts and platforms, pursue multi-channel networks (MCNs) that mass-produce personas across account “matrices”, keep such videos out of rankings and recommendations, and close the worst accounts.
Why a Mandatory Tag Can Become a Shield
The campaign sits on top of the Measures for Labelling AI-Generated Synthetic Content, issued by the CAC, the Ministry of Industry and Information Technology, the Ministry of Public Security and the National Radio and Television Administration, dated 7 March 2025 and in force since 1 September 2025 (in Chinese; translation by AI & GPP). They require visible labels and hidden metadata labels on AI content, and oblige distribution platforms to tag material as confirmed, declared or suspected AI content (Article 6). A mandatory national standard on labelling methods took effect the same day.
The Measures already bar improper labelling that harms others’ rights (Article 10). The new notice extends that logic to the platforms themselves: once a video carries an AI tag, the duty to review it remains. Scale matters. By 31 August 2026, according to the CAC’s filing notice of 14 September, 1,112 generative AI services had completed national filing and 731 apps or features built on them had been registered locally. Each must display the model it uses and its filing number.
Where Beijing’s agent framework governs how AI systems behave, this campaign targets people using AI to deceive.
How Other Labelling Regimes Compare
The table below, built from each jurisdiction’s own legal text or official guidance, asks one question: when synthetic content impersonates a real person, does disclosing it make it lawful?
| Jurisdiction | Instrument | Who discloses | Does a label cure impersonation? |
|---|---|---|---|
| China | Labelling Measures (in force 1 Sep 2025); CAC notice of 30 Sep 2026 | AI providers, platforms, users | No: improper labelling barred (Art. 10); AI or fiction tags no defence for fake personas |
| EU | AI Act Art. 50 (applies from 2 Aug 2026) | Providers (machine-readable marks); deployers (deepfakes, unreviewed public-interest text) | Disclosure duty only; harmful manipulation and deception banned separately since Feb 2025 |
| India | IT Rules 2021, amended 10 Feb 2026 | Intermediaries offering synthetic-content tools; large platforms verify user declarations | No: tools must block content falsely depicting a person in a way likely to deceive (Rule 3(3)(a)(i)) |
| Brazil | TSE Resolution 23.610/2019, arts 9º-B, 9º-C | Whoever is responsible for campaign propaganda | No: campaign deepfakes banned even if authorised; around the vote, AI content of candidates banned “mesmo que rotulados” |
In the EU, the Commission’s guidance on Article 50 frames the duties as disclosure, enforced by national market surveillance authorities and the AI Office. Its AI Act overview lists harmful AI-based manipulation and deception among practices banned since February 2025, and the Commission said it would begin enforcing them from 2 August.
India’s amended Intermediary Guidelines put deceptive impersonation in the block column and label everything else, as this issue’s India report sets out. Brazil’s electoral court has drawn the sharpest line: Resolution 23.610 bans campaign deepfakes of a living, dead or fictitious person “ainda que mediante autorização” (in Portuguese; translation by AI & GPP), a two-track system examined in this issue’s Brazil report.
A disclosure regime that is silent on impersonation invites the fraudster to comply.
What Decision-Makers Should Watch
China’s notice is a campaign, not a statute, and names no specific sanction beyond closing the worst accounts. Watch whether the CAC publishes enforcement figures when the two months end. Elsewhere, regulators writing guidance under labelling rules should state which harms a label cannot cure, above all impersonating real people for money or votes, and keep a platform duty to review that survives the tag.
Sources
- 中央网信办部署开展”清朗·整治短视频虚假人设乱象”专项行动 — Cyberspace Administration of China — 30 September 2026; (in Chinese; translation by AI & GPP)
- 人工智能生成合成内容标识办法 (Measures for Labelling AI-Generated Synthetic Content) — Cyberspace Administration of China — 14 March 2025 (dated 7 March 2025); (in Chinese; translation by AI & GPP)
- 四部门联合发布《人工智能生成合成内容标识办法》 — Cyberspace Administration of China — 14 March 2025; (in Chinese; translation by AI & GPP)
- 关于发布生成式人工智能服务已备案信息的公告（2026年7月至8月） — Cyberspace Administration of China — 14 September 2026; (in Chinese; translation by AI & GPP)
- Guidelines on transparency obligations for providers and deployers of certain AI systems — European Commission — undated policy page; last updated 6 August 2026
- AI Act (Regulatory framework for AI) — European Commission — undated policy page, accessed 11 October 2026
- Commission starts enforcing AI Act rules and new transparency requirements on 2 August — European Commission — 31 July 2026
- The Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021 [updated as on 10.02.2026] — Ministry of Electronics and Information Technology — updated as on 10 February 2026; (PDF)
- Resolução nº 23.610, de 18 de dezembro de 2019 (compiled text) — Tribunal Superior Eleitoral — as amended to 2026; (in Portuguese; translation by AI & GPP)
Discussion
No comments yet. Start the conversation.
