China Court Jails Six AI Companion App Developers Over Explicit Output

A court in eastern China has sentenced six developers of an AI companion app to between three years and one month and three years and ten months in prison, ruling that the people who built and jailbroke the service were the legal “makers” of the explicit material it produced, not the users who typed the prompts.

The case, heard by the Yuyao People’s Court in Ningbo, Zhejiang province, ended in June, but its reasoning has only now been laid out in detail. Xinhua republished a Legal Daily account of the judgment on November 12, and the prosecutors’ newspaper Procuratorate Daily (jcrb.com) ran a follow-up on November 13. Each defendant was also fined. According to Xinhua, none of them appealed and the verdict is final.

What the app did

According to the Supreme People’s Procuratorate’s account, the project was started in July 2024 by a man identified only by the surname Zheng, a former AI programmer at a large Chinese game company, after he came across an overseas AI sex chatbot. He recruited a former colleague and people he met online to form a six-person team.

Their service, named by prosecutors as “OchatAI_bot”, went live on an overseas platform in January 2025. Prosecutors say it ran on a foreign company’s large language model, with rewritten system prompts used to break through the model’s built-in sexual-content filters. Users picked character cards to steer the chat, then spent paid credits to generate matching images on an image model the team ran on rented GPU servers. A later update turned still images into animated GIFs, and the team ran “character card” contests in overseas online groups, rewarding entries with credits.

Investigators say the app generated more than 370,000 still images and more than 6,000 animated ones and took in about $24,000 (more than 170,000 yuan) in top-ups. In a sample of 12,500 stills plus every animation, more than 9,000 stills and more than 5,000 animations were classed as obscene, some depicting sexual violence, according to prosecutors. Police opened the case in July 2025 and arrested all six in September 2025.

“Users made it” did not work

The defendants argued that the app was meant for emotional companionship, that users were the real creators of the content, and that running it abroad was legal. The court rejected all three points. It found that users could only act inside a framework the team had designed, that the team had “substantive control” over what the model produced, and that their work was therefore directly linked to the obscene output. Prosecutors added that internal chats showed the team had fixed payment problems specifically so users in mainland China could pay.

According to Xinhua, the trial judge argued that “technical neutrality” cannot protect a developer who deliberately strips a model’s safety layer. The ruling follows a September 7 opinion from China’s Supreme People’s Court that using AI to make, sell or spread obscene material can be prosecuted as a crime, China News Service reported.

Why it matters

Pornography is illegal in China, so this is not a template for countries where adult content is lawful. But the core question is the same one Western courts now face: who is responsible for what a jailbroken or loosely filtered model generates? In US lawsuits over Grok, xAI has argued that users, not the company, created the images, according to Bloomberg. The Yuyao court took the opposite view for a service built around adult output.

For operators of AI companion and girlfriend apps, the case suggests that a system prompt written to defeat a provider’s filters, plus character cards and paid image credits, can be treated as evidence of intent. For background on how other governments are drawing these lines, see our coverage of the EU AI Act ban on nudify tools and the UK’s new law on realistic AI-generated porn.