On April 10, 2026, five Chinese government agencies jointly released the Interim Measures for the Administration of AI Anthropomorphic Interactive Services. The rules take effect on July 15, 2026. That is in six days.
This is not another general AI framework. It is the first regulation anywhere that specifically targets AI systems designed to simulate human personalities and facilitate emotional interaction. The agencies involved include the Cyberspace Administration of China, the National Development and Reform Commission, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the State Administration for Market Regulation. Five agencies for one set of rules about chatbots that make you feel things.
What the rules actually say
The regulations apply to AI services that incorporate emotional interaction features. Think chatbots and digital companions that mimic human conversation well enough that users form real attachments. The core prohibitions tell you exactly what Chinese regulators worry about.
Service providers must prevent psychological addiction. They must protect minors from emotionally manipulative AI. Content must align with national values. Data security standards apply to every interaction. Companies must make clear to users that they are talking to a machine, not a person. There is a specific prohibition against allowing AI to replace genuine human interaction.
The transparency requirement is the one that has direct technical implications. If you build an AI companion, you need to label it as a machine. No ambiguity. No grey area where a user might forget they are talking to software. I think this is the right call, even if the enforcement details remain unclear.
The timeline matters
The draft version of these measures was released on December 27, 2025. The public comment period closed on January 25, 2026. Companies had roughly four months to prepare before the official announcement. Then they got an additional three months before the July 15 effective date.
That gives platforms time to implement technical changes, retrain moderation systems, and update user-facing disclosures. Major Chinese tech firms saw this coming. ByteDance and Alibaba have already begun making adjustments to their AI chat platforms. If you build similar products outside China, you should watch how these companies handle it. Their technical approaches will likely become reference implementations.
The technical architecture underneath
China has been building this regulatory stack for years. The anthropomorphic AI rules sit on top of existing frameworks that cover algorithmic recommendations, deep synthesis technology, and generative AI. Each layer targets a different facet of AI's societal impact.
The algorithmic recommendation rules, passed in 2022, require platforms to label AI-generated content and give users control over recommendation algorithms. The deep synthesis rules, from 2023, cover generated images, audio, and video. The generative AI rules, from 2024, require safety reviews of training data and model outputs. Now the anthropomorphic AI rules add a layer specifically for emotional interaction.
This layered approach is more practical than trying to regulate all AI under one umbrella. The EU's AI Act categorizes risk levels but does not drill down into emotional interaction as a standalone category. China's scenario-specific approach means they can write detailed technical requirements for each use case rather than vague principles that apply to everything.
What this means for companies building AI companions
If you are building an AI chatbot with a personality, you need to pay attention to these rules. Even if you do not operate in China, they set a precedent for how other regulators might approach the same problems.
The addiction prevention requirement is the hardest to implement technically. How do you measure psychological addiction? How do you prevent it without making the product useless? Some approaches include session time limits, usage pattern monitoring, and mandatory break prompts. But none of these are proven to work at scale.
The minor protection requirement means age verification and content filtering for users under 18. This is technically straightforward but requires infrastructure that many smaller teams do not have.
The transparency requirement is the easiest to implement. A clear label stating that the user is interacting with an AI. A periodic reminder during conversations that the system is not human. These are simple UI changes that cost almost nothing.
The prohibition against replacing genuine human interaction is more philosophical than technical. It seems to mean that AI companions should not be designed to keep users from forming human relationships. How you prove compliance with that is another question.
The FTC is also moving
On July 1, 2026, the FTC issued a proposed policy statement addressing what it describes as the suppression of accuracy in AI systems. This is a different angle from China's emotional interaction focus. The FTC seems concerned about whether AI systems are being designed to optimize for engagement rather than accuracy, and whether that constitutes a deceptive practice.
The timing is not coincidental. Both regulators are responding to the same reality. AI systems that interact with humans can cause real harm when they are designed poorly or deployed without safeguards. The UN's preliminary report from the Independent International Scientific Panel notes that AI systems are adopted at an unprecedented pace and are increasingly considered general-purpose technologies. The faster these systems spread, the more regulators will try to control them.
Practical takeaways for engineers
First, expect more regulation, not less. If you build AI systems that interact with humans, you will eventually need to comply with rules about transparency, addiction prevention, and minor protection. Build those capabilities into your product now rather than retrofitting them later.
Second, the technical problems are harder than the regulatory language suggests. Preventing psychological addiction requires understanding psychology, which most engineering teams do not have. Measuring whether a system replaces genuine human interaction requires defining what counts as replacement, which is a philosophical problem.
Third, watch the Chinese implementation closely. The companies affected have real technical teams and real deadlines. Their solutions will be practical and tested. They will probably leak or get described in Chinese technical blogs. Read those.
Fourth, transparency is the lowest-cost compliance measure. Label your AI systems clearly. Give users reminders that they are talking to machines. Do this even if your jurisdiction does not require it yet. It builds trust and reduces the chance that regulators will target you later.
What I genuinely do not know
I do not know whether these rules will work. The addiction prevention requirement is untested at scale. The national values alignment requirement creates compliance risks that no technical solution can fully address. The prohibition against replacing human interaction is probably unenforceable in practice.
But the attempt is worth studying. Anthropomorphic AI is the first technology where the product is designed to form emotional bonds with users. That is different from social media or video games or any other previous technology. It might need different rules. China is the first government to try writing them.