Phase two of the CAC “Qinglang” AI campaign: 5.61 million items of unlawful content removed and 2,400-plus apps handled
Subject: The CAC “Qinglang: Rectifying Misuse of AI Applications” campaign
What happened
The CAC ran phase two of its campaign against misuse of AI applications, focused on using AI to produce and publish false information, spread violent or vulgar content, impersonate others and harm the rights of minors. It directed provincial cyberspace authorities to push key rectification tasks and required platforms to clear AI-generated unlawful and harmful content and to deal with offending accounts and mobile applications.
How it was handled
The notice lists seven categories of typical problems: using AI to rewrite literary classics into low-quality content; using AI to produce and publish false information (including AI digital humans making false marketing claims); using face- and voice-swapping to impersonate public figures; mass-publishing AI-generated violent or vulgar content; publishing AI-generated content that harms minors; using AI "auto-pilot" tools to run inauthentic-engagement campaigns; and non-compliant AI products, services and applications (cover images, descriptions and generated dialogue of multiple agents containing obscene and pornographic material and other content harmful to minors) — for that last category the notice states that operators refusing to rectify are dealt with under the law, with warnings, fines, delisting or shutdown applied through the applicable procedure. Platforms were required to optimise multimodal detection models and dynamically expand face, voiceprint and violation sample libraries; generative AI platforms to tighten review of source training corpora, restrict non-compliant outputs and implement labelling of AI-generated and synthetic content; and app stores to set developer admission and listing requirements and to re-test apps already listed.
Outcome
As at the notice, the campaign had removed more than 5.61 million items of unlawful or non-compliant content, acted against more than 49,000 accounts, and handled more than 2,400 non-compliant websites and applications; platforms had issued 46 governance announcements. Account names in the notice are partially masked by the regulator.
What it means for Chinese exporters
- · China governs AI content by squeezing app stores, platforms and accounts simultaneously. Chinese AI apps expanding abroad usually also distribute domestically: being delisted at home directly affects overseas fundraising and partner diligence, so domestic labelling compliance and overseas compliance should be one engineering effort, not two tracks.
- · Explicit and implicit labelling has to survive in the actual product output. The notice singles out implementing labelling of AI-generated and synthetic content — whether exported files keep the label, whether it survives re-editing, and whether metadata is stripped are all technical facts that can be spot-checked.
- · Agent cover images, descriptions and personas are in scope, not just dialogue output. When shipping character-based products, apply the same review standards and takedown timelines to user-generated personas as to the main content.
- · AI digital humans making false marketing claims is called out as its own category. Companies marketing with digital humans need an evidence chain covering disclosure of the digital human, script review and product claims — which points the same way as the US FTC’s substantiation expectations, so one cross-jurisdiction evidence pack can serve both.
- · Publishing cumulative totals shows enforcement has moved to routine spot-checking. Budget compliance as a continuously operating review-and-record mechanism rather than a one-off remediation project.
Related regulations
Sources (1)
- CAC advances phase two of the “Qinglang” campaign against misuse of AI applicationsOffice of the Central Cyberspace Affairs Commission (CAC) · published 2026-09-02 · retrieved 2026-09-12
Content as of 2026-09-12 · corpus 2026.09|Compiled from public official sources with links and review dates for compliance decision support. Rules change; have licensed professionals review before external disclosure, regulatory filing or contractual commitments.