A named rule, not a general one, requires consent for a cloned face

The figure or the record

The Cyberspace Administration of China, the Ministry of Industry and Information Technology and the Ministry of Public Security jointly issued Order No. 12, the Provisions on the Administration of Deep Synthesis Internet Information Services (互联网信息服务深度合成管理规定), on 25 November 2022, after review at the CAC's 21st office meeting on 3 November 2022. The text itself, read here in the original Chinese, states in Article 25 that the Provisions took effect on 10 January 2023, a seven-week gap between signing and application that gave platforms time to adjust. The rules cover any organisation or person applying 'deep synthesis' technology, defined in Article 23 to include face generation, face-swapping, voice synthesis and cloning, and full three-dimensional digital-human reconstruction.

What the documents establish

Article 14 is the consent clause a generated performer's rights rest on: a provider or technical supporter offering a tool that edits face or voice biometric data must prompt the user to inform the edited individual, as required by law, and obtain that individual's separate consent, a heightened standard distinct from ordinary agreement to a platform's general terms. Articles 16 and 17 set the labelling duty, requiring a technical mark on generated or edited output and, for categories likely to cause public confusion, including face generation, face-swapping, voice synthesis and 'immersive simulated scenes', a prominent on-content label. Article 18 separately bars anyone from deleting, tampering with or hiding that label by technical means, which converts labelling from a courtesy into an enforceable duty running against the end user, not only the platform.

Scope and caveats

The Provisions regulate deep-synthesis services offered inside mainland China; they do not themselves set the level of a licensing fee, and they leave the underlying civil right they protect to the Civil Code's portrait and voice provisions. 'Separate consent' is not defined down to a specific document format, so a marketplace's own terms are what actually operationalise it for a given transaction. The rule applies to providers and 'technical supporters' as distinct roles, meaning a tool vendor can carry obligations even when it never contracts directly with the person whose face or voice is synthesised.

What to watch

Editorially, the useful signal is whether a specific enforcement action or app-store removal cites Article 14's separate-consent duty or Article 17's labelling duty by name, since that would show which obligation regulators are actually testing against real vertical-drama production rather than treating as aspirational text.

• Does a platform's consent flow ask separately for face or voice editing, or bundle it into a single general terms-of-service acceptance?
• Is the required AI-generation label present on the delivered output, or only promised in the contract text?
• Who is named as the 'technical supporter' in a dispute, and does that role carry its own registration duty under Article 19?

A production or marketplace that says it complies with Chinese deep-synthesis rules should be judged against these named articles, not against a general claim of following 'AI regulation', since the Provisions specify who owes which duty and when it started applying.