On April 12, CCTV's "Financial Investigation" exposed an AI "pornography-creation" industry chain: 9.9 yuan for prompts, 19.9 yuan to turn an image into a video.
On June 22, Xinhua News Agency disclosed a case in Jiaozhou, Shandong, in which a gang used AI-synthesized dynamic facial videos to bypass real-name verification and had sold more than 80,000 accounts.
On June 29, the Nanshan Branch of the Shenzhen Public Security Bureau announced that Luo, a self-media practitioner, had been criminally detained for using AI tools to fabricate videos and spread rumors about Dongpeng Beverage.
The Ministry of Public Security's position in July was that a special campaign against "AI face-swapping" fraud had solved 79 cases and arrested 515 criminal suspects; in the first half of this year, more than 170 cases of AI-generated online rumors were investigated, and more than 190 people were held accountable.
Behind these figures is a complete set of schemes that has already been put into operation. Below, they are listed in three groups according to their distance from the victim's wallet.
Those that take money directly from people
Impersonating relatives, friends and supervisors to ask for money is currently the most common typeVoice cloning now requires only a few seconds of public audio to replicate a voice, at a cost as low as a few dozen yuan; a casual short video on a social platform is enough. The fraudster first buys family relationship and workplace information from a social-engineering database, then creates a voice clone, fabricates an emergency situation, and targets elderly people, students or corporate finance staff. The characterization of such cases is not disputed: it is the crime of fraud; those who provide voice-cloning tools or help prepare materials commit the crime of aiding information network criminal activities.
The "AI refined chatting" in Southeast Asian compounds is the industrialized version of the same thingReuters interviewed Omondi, a young Kenyan man who escaped from Myanmar's KK Park. He said ChatGPT is the tool scammers use most often, to generate conversations with an American tone, impersonating a Texas rancher or an Alabama soybean grower, and to answer in real time the cryptocurrency and real estate questions victims raise. C4ADS, a Washington security research organization, identified from materials leaked from the compounds this year two software platforms that call ChatGPT and Gemini at their core; their functions include automatically generating replies, driving persona role-playing bots, embedding real-time translation in more than 100 languages, and incidentally tracking each worker's performance. The record of an Indian man trafficked into a compound was that in one shift he used dozens of fake accounts to chat with more than 100 people simultaneously, and in one month targeted about 50,000 people from at least 17 countries.
AI impersonation of celebrities for livestream selling is another lineAt the end of 2024, the image of Dr. Zhang Wenhong was AI-synthesized to promote protein bars, and 1,180 of those protein bars were sold. In May this year, the Cybersecurity Administration of the Ministry of Public Security announced a case in which Xing, a netizen in Datong, Shanxi, without authorization used the image of Chinese Kuomintang Chairman Cheng Li-wun to generate an AI digital human for livestream selling, and was administratively detained by the local public security organ, characterized as obtaining money by deception by stealing or falsely using another person's identity. Previously, the Haidian District Market Supervision Administration in Beijing investigated and punished a company that used AI to forge the image of CCTV host Li Zimeng to endorse "deep-sea polyene fish oil," and imposed an administrative penalty under the Advertising Law. The cloning of the voices of Olympic champions Quan Hongchan, Sun Yingsha and Wang Chuqin to sell free-range eggs is also the same category.
The legal characterization of such conduct shifts along a spectrum: using a celebrity's image merely to attract traffic without obtaining money is a civil liability for infringement of portrait rights and reputation rights, plus public security administrative penalties; fabricating facts to obtain property is the crime of fraud; and where the goods promoted are themselves counterfeit, the crimes of producing and selling substandard products are also added.
Face-swap extortion has also been on the rise this yearTrend Micro's annual report identified two categories of tools, one being "undressing" applications and the other being AI-generated sexualized imagery, which are driving up the number of sexual extortion and face-swap sexual extortion cases. The modus operandi is short: scrape a few photos from social media accounts, generate indecent content, and then use it to demand money or more imagery. The conduct is characterized as the crime of extortion; where the content has been disseminated, the crime of disseminating obscene materials is also added.
One person, dozens of identities
The several types who sell tools and raw materials to the people above
The second group is the most stable-shipping part of this industrial chain.
Sellers of real-name accounts rank firstThe Jiaozhou case ran the full chain: on an overseas chat tool, they bought citizens' personal information at 5 to 10 yuan per entry, including names, ID numbers and ID photos; used overseas AI tools to enlarge and restore the photos, then synthesized dynamic videos of "blinking" and "shaking the head"; when registering social accounts that required facial authentication, instead of invoking the camera, they used a "virtual camera" software to feed the prepared videos directly in; the real-name accounts that passed verification were sold downstream at 30 to 50 yuan each, and buyers used them to post information for overseas gambling and pornography websites. The case-handling police explained that real-name accounts have higher information-publishing privileges than non-real-name accounts, which is why they are valuable. On this chain, both the sellers and buyers of information commit the crime of infringing citizens' personal information; faces are biometric information, and Guiding Case No. 192 of the Supreme People's Court has already made clear that it is protected by criminal law.
Sellers of undressing software and prompts rank secondIn that CCTV program, the reporter paid 9.9 yuan for advanced prompts, 19.9 yuan to activate image-to-video, and 29.9 yuan for the function to "strip" images and videos; generating a six-second video consumed platform coins equivalent to less than 0.2 yuan. This business already has two judgments: the Sun case, first-instance on May 9 by the Beijing Haidian District Court, in which he packaged his self-developed one-click undressing process into a tutorial and opened paid groups, with membership fees rising from 38 yuan to 168 yuan, accumulating 27,000 community members and illegal gains of more than 100,000 yuan; he was convicted of the crime of illegally using information networks, sentenced to one year and six months' imprisonment and fined 20,000 yuan; in another case, Zhang sold "AI one-click undressing" software at 168 yuan, made more than 100,000 yuan in profits, and was sentenced. Earlier still was the Yu case, Hangzhou Internet Court (2023) Zhe 0192 Min Chu No. 4563, in which he collected face information of more than 100 people to make face-swap obscene videos, set up groups and collected group fees, and also sold software, taught tutorials and took face-swap customization orders; he was prosecuted criminally for the crime of producing and disseminating obscene materials for profit, and in the civil public interest litigation was ordered to pay 60,000 yuan in compensation and to publish an apology in Legal Daily.
Those who build customized systems for the parks rank third, and also quote the highest prices among this groupTMTPost interviewed a technology broker who regularly travels among China, Malaysia and Singapore; the "AI refined chat system" he supplied to the parks was sold by several hundred teams in Southeast Asia in earlier years, at prices ranging from 100,000 to 500,000 yuan; later it was upgraded to park-exclusive large models, trained on the parks' own data through cleaning and annotation, selling at USD 180,000 per set, and as an agent he earned a USD 50,000 commission on each set sold. In his own words, when new technology emerges, the first to adopt it will always be the industry closest to money. This role is not legally ambiguous: where a person knows the other party will use it to commit telecom and online fraud and still provides technical support, at the least it is the crime of aiding information network criminal activities; where there is a common intent, it is directly handled as an accomplice to fraud.
Another type that has just emerged involves removing AI labels and watermarks for others. Many people see this as a trivial favor, but after the Measures for Labeling AI-Generated Synthetic Content took effect on September 1, 2025, it has been explicitly written into the prohibited provisions, as will be discussed later.
The shelves along this chain
The kinds that do not directly defraud money but levy an IQ tax
This group is closest to ordinary people and is most easily mistaken for legitimate business.
The largest in number are AI courses and "AI automated product-selling" trainingIn the complaint investigation published by CNR in March, one company's cooperation packages were divided into three tiers of RMB 3,760, RMB 5,210 and RMB 9,510, promoted as "zero followers, zero foundation" and "log in every day, withdraw commissions and earn lying down"; after signing, trainees were pulled into a "dedicated service group" in which they were the only member and everyone else was a company employee. One consumer spent more than half a year and earned only about a hundred yuan in commissions; when seeking a refund, they were told that, taking the RMB 9,510 package as an example, RMB 6,807 in service fees would be deducted. Another person who paid RMB 2,880 later summed it up by saying that in the eyes of such companies, AI is a tool, while the sickle is their industry.
The same script has already spread down to counties and rural areas. Southern Weekly reported a case in April in which nearly 2,000 middle-aged and elderly trainees were attracted by short videos saying "you can make money even with zero foundation in the AI era," moving from a RMB 1 course to a RMB 2,800 partner program and then to offline high-priced courses costing from over a hundred thousand to several hundred thousand yuan, with one person having invested more than RMB 370,000 cumulatively. The registration form stated referral rebate ratios ranging from 25% to 50% for recommended transactions.
The characterization of this category must be viewed separately. Exaggerated promotion and promises of returns that simply cannot be achieved constitute false advertising under civil law, and a claim for treble damages may be asserted under the Law on the Protection of Consumer Rights and Interests; where, at the time of taking money, there was neither the ability nor the willingness to perform, the nature slides toward the crime of fraud; and where the hierarchy and rebate structure are established, it may also touch on the crime of organizing or leading pyramid selling activities.
The last type is AI rumor-mongering to short a stock and corporate extortionThe Eastroc Beverage case is the latest sample: on June 22, an AI-forged "internal dinner" video spread on short-video platforms, in which the face of founder Lin Muqin was transplanted onto someone else, paired with the line "I don't usually drink this"; from June 22 to 26, the company's A-shares and H-shares shed more than RMB 7 billion in combined market value; on June 27, the party involved released the original video to clarify, and on June 29, police reported that the rumor-monger, Luo, was criminally detained for using AI tools to fabricate false information and attract attention and traffic for profit, and the company's share price hit the daily limit that day. Previously, Nongfu Spring and Wahaha had both encountered similar content. The industry saying is that this chain has already been divided: some collect corporate materials, some handle model generation, and some operate account matrices for distribution.
Such content targeting enterprises may simultaneously violate the crime of damaging commercial reputation and product reputation and the crime of picking quarrels and provoking trouble; fabricating and disseminating false information that affects securities trading carries a separate charge; and those who conveniently demand a "post-deletion fee" from enterprises commit the crime of extortion.
The lines these schemes cross are, in fact, just a few
Gathering the dozen or so categories above, there are not many charges; the same few recur.
Using AI to defraud money directly is the crime of fraud. Where AI is used to fabricate identities to obtain property, the technology is merely the means; the charge is no different from making fraudulent phone calls to obtain money thirty years ago.
Providing tools, accounts or technical support to fraudsters constitutes the crime of assisting information network criminal activities; setting up groups to sell tutorials or disseminate illegal information constitutes the crime of illegally using information networks. This is what is noteworthy in the Sun case: he was convicted not of producing obscene materials but of illegally using information networks, on the ground that he had established communication groups used to impart illegal methods. This charge does not require that you act with your own hands.
Involvement with faces and voices constitutes the crime of infringing citizens' personal information. The threshold for criminal liability is lower than most people assume: under the judicial interpretations of the Supreme People's Court and the Supreme People's Procuratorate, 500 items of such information constitutes "serious circumstances," and illegal gains of RMB 5,000 or more likewise constitute "serious circumstances."
Producing pornographic content constitutes the crime of producing, reproducing, publishing, selling or disseminating obscene materials for profit. In the first AI chat application pornography case, tried at first instance last September, the two founders were sentenced to four years and one year and six months respectively, and fined RMB 4 million and RMB 200,000; both have appealed.
Conduct and the corresponding charges
Three points are most easily misunderstood and warrant separate discussion.
First, "I only do the technology and do not ask about the use" does not shield you. "Knowledge" under the Criminal Law may be inferred, and abnormal unit prices, requests for no watermark, requests for bulk account generation, and settlement through personal accounts are all common bases in judicial practice for inferring knowledge. When the client is arrested, the person doing the technology often goes in as well.
Second, those who sell tools are not much safer than those who use them. Of those sentenced in the three judgments above, two sold software and tutorials, and one both produced content and sold tools. The most profitable party in this chain is the most upstream, but the most upstream is equally within the reach of the Criminal Law.
Third, AI labeling is not optional. The Measures for Labeling AI-Generated Synthetic Content and the supporting mandatory national standards took effect simultaneously on September 1, 2025, and Article 10 thereof provides:
Where a user publishes generated synthetic content through a network information content dissemination service, the user shall proactively declare it and use the labeling function provided by the service provider to label it.
No organization or individual may maliciously delete, tamper with, forge or conceal the generated synthetic content labels provided for in these Measures, may provide tools or services for others to carry out the aforementioned malicious acts, or may harm the lawful rights and interests of others through improper labeling means.
The second half of the second paragraph directly brings "removing watermarks for clients" and "selling de-identification plug-ins" within the scope of regulation. The amended Cybersecurity Law took effect on January 1 of this year, establishing for the first time at the legislative level a liability framework for AI security risks. In this direction, the gaps are being filled.
Back to the question in the title. The cases from the past six months do show that AI is far more efficient at generating value in the hands of the gray and black markets than in legitimate businesses, because it saves three costs: compliance, raw materials, and trial and error. But putting the illegal gains and sentences in these judgments together reveals something else: the 14 members of the Jiaozhou group split more than 600,000 yuan; Sun earned more than 100,000 yuan and received one year and six months; and the judgment described Yu Mou-long's illegal gains as "a total profit of several thousand yuan." Those who truly made money in this chain were at the layer selling a model for USD 180,000 per set, not the layer doing the hands-on work.
For those doing legitimate business, there is really only one thing worth remembering: what separates you from these judgments is not "whether you want to do bad things," but several very specific actions. Keep a record of the material authorization chain; sign authorizations for likeness and voice separately, specifying the purpose and term; retain service logs for no less than six months as required by the Measures; in contracts for undertaking commissioned work, clearly state use restrictions and the client's commitment regarding use; do not accept orders priced significantly below market; do not remove labels, and if a client asks you to remove them, treat that request itself as a risk signal.
Those with relevant legal needs may speak with the author.
Reference Sources
- China Media Group, "Financial Investigation," investigation into the AI "pornography creation" industrial chain, April 12, 2026
- Xinhua News Agency, "Xinhua Investigation | Exposing the Black Industrial Chain of 'AI Face-Swapping' Fraud," June 22, 2026
- Police Situation Report of the Nanshan Branch of the Shenzhen Public Security Bureau, June 29, 2026; related reports by Securities Times and 21st Century Business Herald
- Notice of the Ministry of Public Security on the progress of the special action against "AI face-swapping" fraud, July 2026
- Jiangsu Procuratorate Network, "Sentenced for Peddling 'AI Undressing Tutorials'!," prosecuted by the Haidian District People's Procuratorate of Beijing, first-instance judgment by the court on May 9, 2026
- National Business Daily, report on Zhang being sentenced for selling "AI one-click undressing" software, July 2026
- Hangzhou Internet Court (2023) Zhe 0192 Min Chu No. 4563 civil judgment, Supreme People's Court Case Database entry No. 2025-07-2-525-001
- Guiding Case No. 192 of the Supreme People's Court: Li Kaixiang's Criminal Case with Incidental Civil Public Interest Litigation for Infringement of Citizens' Personal Information
- Reuters interview with a person who escaped from the KK Park in Myanmar; Lianhe Zaobao Chinese-language report, September 15, 2025
- Analysis by C4ADS, a Washington-based security research institution, regarding AI software platforms and device networking records in Myanmar telecom fraud parks, July 2026
- TMTPost, "Exposing Southeast Asia's AI Legions: Dedicated to Scam Parks, Telecom Fraud Efficiency Soars Tenfold"
- Cybersecurity Bureau of the Ministry of Public Security, notice on administrative detention of Xing Mou in Datong, Shanxi, for using AI to impersonate a celebrity in livestream selling, May 7, 2026
- CCTV.com "News+", report on the incident of AI face-swapping involving Zhang Wenhong to promote protein bars, December 18, 2024
- Xinhuanet, "Regulating the Chaos of AI Celebrity Endorsement in Livestream Selling: E-commerce Livestreaming Launches a Portrait 'Defense War'", May 14, 2026
- CNR, Woodpecker Consumer Complaints | Investigation into the Chaos of "AI Automated Sales" Training, March 15, 2026
- Southern Weekly, "One Yuan for Traffic, Exorbitant Prices for Courses: 'AI Financial Quotient Courses' Target Rural Elderly", April 8, 2026
- Trend Micro, "2026 Fraud Trend Forecast"
- Cyberspace Administration of China and three other departments, "Measures for Labeling AI-Generated Synthetic Content", Guoxinban Tongzi [2025] No. 2
- Supreme People's Court and Supreme People's Procuratorate, "Interpretation on Several Issues Concerning the Application of Law in Handling Criminal Cases of Infringing Citizens' Personal Information"
