What criminal risks should AI entrepreneurs be aware of when going global? Analysis of 5 popular sectors.

Author: Lawyer Shao Shiwei

From a macro perspective, AI exports are experiencing a period of rapid growth. In the first half of 2026, China's AI-related product exports accounted for 22.7% of total exports, contributing 8.6 percentage points to overall export growth. AI going global is already an industry consensus.

However, for AI entrepreneurs, going global doesn't mean they only need to consider overseas markets and local regulations.

On September 7, 2026, the Supreme People's Court issued the "Opinions on Legally Handling Disputes Involving Artificial Intelligence," explicitly including acts such as using artificial intelligence to commit fraud, infringe on citizens' personal information, and produce and disseminate obscene materials within the scope of criminal prosecution.

In the same month, the first case in China involving AI going overseas and being involved in pornography was publicly sentenced (➡️《"Breaking the AI ​​Model to Profit from Pornography? Court: It Constitutes the Crime of Producing Obscene Materials for Profit!") - Developer Zheng was once an AI programmer at a major domestic game company. His "OchatAi_bot" chatbot was launched on an overseas platform in January 2025. After discovering illegal clues, the public security organs arrested all the people involved in the case within two months. The Yuyao Court of Zhejiang Province sentenced six people to more than three years in prison for the crime of producing obscene materials for profit.

(Software involved in the case)

This article aims to discuss: Facing the industry trend and market opportunities of AI going global, where exactly is the boundary between growth and compliance for AI entrepreneurs?

1. Why are more and more AI entrepreneurs choosing to go global?

In the past, when discussing going global, the logic was often "the domestic market is too competitive, so we go overseas to find growth." But now the driving forces behind AI going global are more complex.

On the one hand, there is a supply mismatch.

Many AI application categories that have already been successfully implemented overseas—video translation, e-commerce product image generation, automated changelog—either lack comparable products in China or are not well-developed. Using proven scenarios to serve the global market has a higher success rate than competing in the red ocean of China. On the other hand, there is a difference in payment structure. IDC data shows that hardware accounts for as much as 74.3% of AI spending in China, while hardware accounts for only 18.0% in Latin America, and software accounts for 60.9%[1]. Overseas markets are more willing to pay for software and solutions rather than buying servers and computing cards. This is a clear entry point for Chinese teams that are good at application layer development. Simply put, the advantages of Chinese AI entrepreneurs are: lower model and computing power costs, proven application scenarios, and small and fast full-stack teams. These advantages point to overseas expansion directions that are mostly concentrated in a few tracks: AI companion social networking, digital humans, AI marketing, AI stock trading, and AI Agent. These sectors have genuine market demand and mature overseas counterparts. However, after product launch, as pressure from user needs, paid conversions, and revenue growth increases, some originally normal product features may be pushed further forward, gradually entering high-risk legal territory. Often, entrepreneurs don't intentionally break the law; rather, the boundaries slowly change through repeated product iterations and business choices. 2. The Evolution of Risks in Popular AI Sectors: Where Did They Go Wrong? Few entrepreneurs in the AI ​​sector initially intend to create illegal products. A more common path is: the team discovers that a certain type of user need has higher retention and a stronger willingness to pay, so they continuously optimize the product around these needs—until the product's core selling point itself crosses legal boundaries. The following analysis, using specific cases, examines the risk evolution paths of several typical sectors.

1. AI Companionship/Social Interaction: From Emotional Value to Paid Pornographic Content

AI companionship is a real and rapidly growing sector. The initial products focused on emotional companionship and role-playing in romantic scenarios. However, after a period of operation, teams discovered a very real data pattern: ordinary chat users would leave after a few conversations, while romantic interaction users had higher retention rates. Furthermore, as the level of explicit dialogue increased, chat duration, points consumption, and recharge rates rose significantly.

At this point, companies faced a business decision: since the more explicit the content, the more profitable it was, should it be made a core selling point? Thus, they began adding more explicit character cards, adjusting system prompts to optimize pornographic dialogue capabilities, and adding image and video generation to higher-level paid packages. The product progressed step by step along the path of "AI companionship → romantic simulation → pornographic chat → paid generation of obscene content."

The first case of domestic AI going overseas involving pornography, mentioned at the beginning of this article, is a typical example.

Developer Zheng's initial motivation was the market opportunity in the AI ​​companionship sector, but the final product evolution led the entire team into the realm of criminal law. Similarly, if AI dating products begin using algorithms to identify users' willingness to pay, shifting the matching results from "dating" to "paid companionship" or even facilitating offline special services, the risk path follows the same logic. Lawyer Shao advises: The real point where risk escalates is not when a user makes an inappropriate request, but when the company begins to proactively optimize its product and design growth strategies around high-risk demands. 2. AI Marketing: From Automated Customer Acquisition to Automated Fraud Infrastructure The core value of AI in marketing is reducing the cost of large-scale personalized communication. However, the same capabilities—mass-generating personas, automating private messaging, customizing scripts based on user profiles, and continuously cultivating trust—become a fraudulent assembly line when used for fraud conversion. In July 2026, OpenAI publicly disclosed that it had blocked a fraud network based in Poipet, Cambodia [2]. The network used ChatGPT to translate messages, generate social media content, and build fake identities to carry out investment fraud and romance fraud according to the standard process of "contact - induce emotions - harvest money". What AI does here is to scale up the whole process of "account nurturing - chat - build trust - conversion" that originally required a lot of manual work. For entrepreneurs of AI marketing SaaS, the dangerous evolution path is: ordinary AI copywriting tools are highly competitive and not expensive, but if the product can register accounts in batches, automatically generate personas, maintain relationships through long-term chat, and accurately convert, some customers are willing to pay much higher prices. In order to continuously add functions such as fake identities, batch account nurturing, and emotional inducement for these indicators, the product may change from a "marketing tool" to the infrastructure of illegal business. Attorney Shao's Tip: The features clients are most willing to pay for may be the high-risk product features. When a team starts differentiating itself around these features, company leaders and product managers need to further assess: Is this feature feasible, and should it be implemented? 3. AI Stock Trading: From Assisted Analysis to Trading for Users AI as an auxiliary investment analysis tool is a normal direction for fintech; however, once AI directly provides users with stock buying and selling advice or specific investment conclusions, it is no longer a "normal product direction," but rather a securities investment consulting business that requires a license. In September 2026, the Supreme People's Procuratorate released typical cases of technical support for intellectual property prosecution cases, including the "DIY Stock Trading Robot" case tried by the Jing'an District Court of Shanghai. (➡️Are AI stock trading software illegal? Are programmers breaking the law by developing such products? Analysis of a real case in Shanghai). This company initially developed stock quantitative SaaS software. Due to the lack of securities qualifications, early users could only perform data analysis and simulated trading; actual trading had to be done on the brokerage platform, resulting in very few users. To overcome its operational difficulties, the company began to add more direct trading capabilities—the product could later automatically buy and sell stocks based on preset conditions, and it bypassed the technical protection measures of the brokerage trading system by purchasing a program to achieve integration. Ultimately, the actual controller was sentenced to seven years and nine months in prison for illegal business operations and copyright infringement. In September 2026, this case was selected by the Supreme People's Procuratorate as a typical case of technical support for intellectual property prosecution. Lawyer Shao's advice: From tools to services, from analysis to decision-making, from assistance to replacement, each step may trigger different regulatory qualification requirements. 4. AI Digital Humans and AI Agents: The Stronger the Technology, the Greater the Risk Exposure What legal risks might AI digital humans face when used for face-swapping, voice-changing, and impersonating real people? AI digital humans have a wide range of normal applications—live-streaming e-commerce, corporate training, and short video production—but the same technology used for face-swapping, voice-changing, and generating false identity content carries criminal risks. In June 2026, the U.S. Department of Justice and the Department of Homeland Security jointly shut down two deepfake websites and coordinated with law enforcement agencies in France and Italy. This shows that some applications are not as simple as AI entrepreneurs and developers mistakenly believe: "They can't be done domestically, but they're fine overseas." Where are the boundaries for AI agents acquiring user data, account keys, and third-party system permissions? Regarding AI Agents, when "fully automated" becomes a core selling point, teams continuously grant the Agent more permissions: reading more data, saving login status, and cross-platform operation. The greater the permissions, the better the product experience, and the more willing users are to pay—but once the Agent's permission design crosses the legal boundaries of data security and computer crime, it may constitute a crime of infringing on citizens' personal information or illegally obtaining data from computer information systems. The Cyberspace Administration of China's normative document released in April this year directly points out the risk scenario of "using intelligent agent technology to steal user data or account keys." Attorney Shao advises: The common characteristic of these two tracks is that improvements in technical indicators (more realistic, more autonomous) and increases in legal risks often go in the same direction. When teams pursue "more lifelike" and "more automated" as product competitiveness, they need to simultaneously assess whether each technological upgrade has crossed into an area requiring compliance review.

3 How can AI entrepreneurs control criminal risks in the early stages of product growth?

Reviewing the above cases, we can see that AI entrepreneurs did not initially intend to create illegal products. The developers in the Yuyao case saw a market opportunity in the AI ​​companion sector, while the founder of the "DIY stock trading robot" initially created a compliant quantitative analysis tool. What truly led the product into the criminal risk zone were seemingly reasonable product decisions made under growth pressure—each step, individually, was about "optimizing customer experience" and "improving conversion rates," but accumulated, the nature of the product fundamentally changed.

This means that compliance work for AI going global cannot wait until product problems arise. Entrepreneurs need to establish several basic checks early in product design and growth decisions:

First, identify how far your product is from the legal red line. Some sectors are naturally closer to risk—products involving content generation, financial transactions, identity simulation, and data collection should have clear boundaries set from day one.

Second, be wary of the "most profitable feature." When a team discovers that a certain type of demand or feature has brought far higher-than-expected paid conversion rates and retention rates, it's precisely at this point that they should stop and examine: Is this growth driver itself compliant? Third, don't assume that "going overseas exempts you from domestic law." The first case of AI-related pornography overseas has proven that AI products operated by Chinese citizens on overseas platforms are also subject to domestic criminal law. For entrepreneurs, opportunities and risks often follow the same path. Making trade-offs between conversion efficiency and compliance boundaries is a challenge every AI entrepreneur must face from day one of product development. [1] IDC China: "Same AI Budget, Different Spending Methods in China and Abroad: Time Difference and Structural Opportunities for Going Global Based on IDC Global AI Spending Guide" [2] OpenAI Official Website: "Disrupting Malicious Uses of AI — Criminal Scam Operation"