On 20 June 2025 the Cyberspace Administration of China published the phase one progress of the Qinglang campaign against the misuse of AI technology. These figures are an important reference point for judging enforcement intensity and priorities.
The enforcement intensity behind the numbers
Phase one handled more than 3,500 non-compliant AI products — mini-programs, applications and agents among them — cleared over 960,000 items of unlawful information and dealt with more than 3,700 accounts. Cyberspace authorities in many localities pursued local governance, and major platforms published their own measures in parallel.
The scale shows that this was not a pilot exercise but a systematic clean-up spanning platforms and products.
Two core problems
The campaign focused on two problems: AI face swapping and voice cloning that infringe the rights of the public, and the absence of AI content labels that misleads the public. The first concerns personality rights, the second the truthfulness of information. See our note on the Measures for Labelling AI-Generated Synthetic Content.
These map onto two directions of governance: controlling the misuse of technology that impersonates others, and controlling an information environment in which true and false are hard to tell apart.
Where phase two points
The announcement confirmed a second phase focused on seven prominent problems, including AI-generated rumour and vulgar content. Governance is moving from clearing what already exists to regulating what is newly created — the emphasis is no longer on the compliance of products already online, but on the continuing compliance of newly generated content.
Compared with earlier campaigns, this one places more weight on detection and labelling capability on the technical side. See our notes on the Qinglang campaign against AI application disorder and on its first phase.
Where companies can start
First, check whether products touch face swapping, voice cloning or missing labels; these are the high-priority risk points.
Second, make labelling a default product configuration, not an optional add-on.
Third, build a record-keeping mechanism for content generation and distribution, in anticipation of verification in later phases. See our notes on the reporting channel for AI application misuse and on Qinglang’s campaign against malicious hype around corporate information.
The rhythm of these campaigns carries a message: AI content compliance is not a one-off rectification but a capability that has to be maintained continuously.
This article is based on the Cyberspace Administration of China’s phase one progress report on the Qinglang campaign against the misuse of AI technology and public information and is for reference only; it does not constitute legal advice.