What was announced, and what was not
On 30 September 2026 the Cyberspace Administration of China (CAC) announced a two-month nationwide campaign against fabricated personas in short videos, MLex reported on 8 October. Its stated target includes impersonation using artificial intelligence, one of five priorities. The MLex page does not list the other four. It also does not describe platform duties or recommendation algorithms. Reports that platforms cannot excuse such content with labels like "fictional" or "AI-generated" come from a secondary tracker. We could not check that against the CAC's own notice, so treat it as unconfirmed. What follows on algorithms is our analysis, not reporting on the notice's text.
The strongest case for the campaign
The regulator's concern is legitimate. Staged hardship content works because it exploits sympathy. In April 2025 the CAC ran a three-month drive against creators who staged scenarios, including people posing as vulnerable delivery workers, inventing elite personas, or fabricating tragic rural stories for attention and profit. Viewers are misled, donations and purchases can follow, and genuinely poor people are made less believable. AI impersonation of public figures adds fraud risk. Any democracy would want a remedy for this, and consumer-protection regulators elsewhere are pursuing similar conduct.
Why the creator is the wrong unit of enforcement
A campaign aimed at accounts treats the persona as the cause. But a fake persona is only profitable if distribution rewards it. Staged conflict and manufactured hardship generate watch time, shares and comments. A recommender trained to maximise engagement will surface them without anyone deciding to.
China has had a legal hook for this since 2022. The Provisions on the Administration of Algorithmic Recommendations for Internet Information Services took effect on 1 March 2022. Providers with public-opinion attributes or social mobilisation capacity must file through the algorithm filing system within ten working days of starting service (Articles 24–26). Article 14 bars using algorithms to fake accounts, likes, comments or reposts, and to over-recommend content or rig rankings. Users must be able to switch recommendations off or use a version not based on personal characteristics (Articles 16–17).
None of that says that a platform must discount traffic from an unverified persona. Nor does it require the filed algorithm to be audited for rewarding manufactured conflict. That gap is where we would expect the campaign's real weight to fall: not on a single creator's account but on how the platforms' filed systems amplify such content. Whether the notice actually says this is something the reported text does not tell us.
The label question
China's Measures for Labeling AI-Generated Synthetic Content, signed 7 March 2025 and effective 1 September 2025, require explicit labels and metadata-embedded implicit labels. Article 6 obliges dissemination platforms to check metadata and flag content as confirmed, possible or suspected AI-generated. Labels therefore disclose how content was made. They do not say whether a persona is true. If the unverified report is right that "fictional" and "AI-generated" tags cannot excuse fabricated personas, the logic is coherent: disclosing that a video is synthetic does not cure a deception about who the speaker is or what happened to them. The rule would work as a deception standard, separate from the labeling regime.
The proportionality problem
The pro-innovation objection is not that fraud should go unpunished. It is that vague standards pushed through campaigns produce over-removal. "Staged" and "false identity" are clear at the extremes. A scripted comedy sketch, a role-play channel or an actor playing a doctor sits closer to the line. In campaign mode, platforms face the incentive to remove first because penalties fall on them, not on the person who loses a video. When the enforcer also sets the ranking priorities, as Article 11 of the 2022 provisions does by requiring mainstream-value content in prominent slots, a platform has little room to design its own trust signals.
A more proportionate design would be narrower. It would define the harm as deception that induces payment, donation or reliance, and treat fictional formats that are plainly framed as fiction as outside the target. It would ask platforms to publish what they changed in ranking, not only how many accounts they removed, and it would give creators a stated route to contest a removal. Verified-credential badges for people claiming professional expertise would do more than takedowns, because they give audiences information without erasing speech.
What to watch
Three things will show whether this campaign targets creators or systems. First, whether the CAC's notice, once checked, names recommendation or traffic mechanisms. Second, whether enforcement reports count algorithm changes alongside account removals. Third, whether the campaign leaves a durable rule or lapses at the end of its two months, as the April 2025 drive did. Campaigns that end without codifying a standard leave creators guessing and platforms over-complying.
The algorithm registry is the more accountable instrument, because it already exists in written law. A campaign that used it openly would be easier to evaluate than one that works through repeated sweeps.