---
title: "A virtual influencer sold contact lenses. The fine was 68,300 yuan."
date: 2026-11-20
category: Policy & Governance
site: NeuroAI
canonical: https://neuroai.site/a/na-policy-ai-cleanup-1120
language: en
---

# A virtual influencer sold contact lenses. The fine was 68,300 yuan.

> China's AI cleanup cleared 5.6 million items and closed 49,000 accounts. A virtual influencer selling contact lenses was fined 68,300 yuan.

An AI-generated heroine in a micro-drama told viewers: I wore these lenses all day and they felt fine.

Securities Times reported that the avatar, named Fang Taozi, was fined. Administrative penalty records published by Credit Guangdong show a fine of 68,300 yuan for fabricating a user experience and publishing an advertisement for a Class III medical device without the required medical advertising review.

For an account that can mass-produce "I tried it myself" claims, is 68,300 yuan expensive or cheap?

## 5.6 million items, 49,000 accounts

On November 19, the Cyberspace Administration of China published typical cases from the second phase of its campaign against AI misuse.

Three numbers appear in the notice: more than 5.61 million pieces of illegal or harmful information cleared, more than 49,000 accounts investigated, and more than 2,400 websites and applications dealt with.

5.61 million divided by 49,000 is about 114.

On average, each account removed took 114 pieces of content with it. That is my division and the two figures may not align precisely, but the scale relationship is clear: accounts are nodes, content is flow, and enforcement starts at the node.

## Seven categories, one motive

The notice lists seven: AI rewriting of classic literature; AI-generated fake disaster and rescue footage; face and voice swapping to impersonate public figures; bulk posting of violent or vulgar material; AI content harming minors; AI-managed accounts used for paid posting; and AI applications whose agent profiles and outputs carried prohibited content.

In five of the seven, the common thread is the same: AI is a tool, and the goal is traffic and money.

That explains why enforcement lands on accounts and applications rather than on models.

## Labelling solves awareness, not truth

Securities Times noted that the measures on labelling AI-generated content took effect last September. In March this year, the market regulator clarified that recommending or endorsing a product through an AI-fabricated persona, without a prompt that the content is AI, may be treated as false advertising where consumers could be misled.

One line from that commentary is worth keeping: labelling something as AI-generated is not the same as being compliant. Labelling addresses awareness; truth determines whether an advertisement stands.

Applied to this case: even if viewers know the face is synthetic, presenting an experience that never happened as fact is still false.

## What a 68,300-yuan fine signals

The fine maps to a Class III device advertisement without review plus a fabricated experience.

For one account, that is real money. Across a chain — scripting, generation, placement, revenue share — it rounds to nothing at each link.

Which is why the fine amount is not the powerful part. The phrase "may be treated as false advertising" is.

It moves AI virtual-influencer selling out of a grey zone and into the scope of advertising law, which is where joint liability and platform review duties acquire a foothold.

The notice closes with a standard line: regulators will keep watching new situations in synthetic content and guard against misuse.

Enforcement usually trails diffusion by six to twelve months. In that window, platforms are the ones holding the line.

## The economics behind the enforcement

Account farming with AI is cheap because the marginal cost of one more video is close to zero. A single operator can run dozens of accounts, each posting synthetic content until a platform notices.

That asymmetry is what the campaign targets. Clearing 5.6 million items and closing 49,000 accounts is a volume response to a volume problem, and it works only if platforms keep detecting new accounts faster than they are created.

The application side is harder. More than 2,400 websites and apps were dealt with, and the notice singles out agent profiles and generated dialogue containing prohibited content. Agents are harder to review than posts: the output depends on the prompt, so the same product can produce compliant and non-compliant answers within a minute.

For brands, the practical lesson is narrower. An avatar can sell, but only if the claim is true and the disclosure is present. The fine is small; the precedent is not.

## Why the cases are published at all

Publishing anonymised cases serves a specific purpose: it tells platforms what to look for and tells operators what the boundary is. Fines settle individual cases; notices set expectations for the next thousand.

*Disclaimer: This article draws on the Cyberspace Administration notice and Securities Times reporting. Calculations are the author's own. It is for reference only and does not constitute investment advice.*

---

Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-policy-ai-cleanup-1120
Free to quote with attribution and a link to the original.
