China's Interim Measures for Anthropomorphic AI Interaction Services took effect on July 15, 2026, restricting AI that simulates emotional human relationships while exempting functional tools like customer service bots and workplace assistants. Arriving days before the EU AI Act's August 2 chatbot-disclosure deadline, it marks a global shift toward mandatory transparency for customer-facing AI.
For two years the regulatory conversation around AI has centered on training data, safety testing, and high-risk use cases. A quieter question is now moving to the front: when a machine talks to a person, does that person have a right to know it is a machine, and how human-like is that machine allowed to seem? Three of the world's largest regulatory blocs are answering at nearly the same moment, and the answers rhyme.
What China's New AI Rules Actually Do
The direct answer: China's Interim Measures for the Administration of Anthropomorphic AI Interaction Services regulate AI services that simulate human personality traits and provide sustained emotional interaction. The rules were jointly issued on April 10, 2026, by the Cyberspace Administration of China alongside four other agencies, and they took effect July 15. Coverage from the South China Morning Post and legal analysis from Bird & Bird describe a framework aimed squarely at AI that mimics human companionship rather than AI that performs tasks.
The scope is narrower than the headlines suggest. According to reporting from AI News, the measures apply to services offering "continuous emotional interaction simulating natural persons' personality traits, thinking patterns, and communication styles." They explicitly exclude intelligent customer service, knowledge question-and-answer tools, workplace assistants, education, and scientific research. Services that launch anthropomorphic functions or cross thresholds of one million registered users or one hundred thousand monthly actives must run security assessments across eight areas and file the reports with provincial regulators. Providers are also barred from offering virtual companion services to minors and must build dedicated minor modes with usage limits.
The effect was immediate and visible. As Decrypt and other outlets reported, ByteDance told users that Doubao's agent function would go offline on July 15, and Alibaba's Qwen said its human-like and user-created agents would stop working ahead of the deadline. When persistent-memory companion agents meet rules requiring anti-addiction friction, mandatory usage notifications, and instant-exit mechanisms, the two designs turn out to be architecturally incompatible.
Why This Matters Well Beyond China
It would be easy to file this as a China-only story about consumer companion apps. That reading misses the pattern. The same principle, that people should know when they are dealing with AI, is landing in Western law on almost the same timeline.
In the European Union, Article 50 of the AI Act requires that providers design systems intended to interact with people so those people are informed they are talking to AI, unless it is obvious to a reasonably well-informed person. The official Article 50 text sets these transparency obligations to apply from August 2, 2026, roughly two weeks after China's rules took effect. Reported penalties for related violations run up to fifteen million euros or three percent of worldwide annual turnover, whichever is higher, as detailed in guidance on the Act's transparency rules. We tracked the moving pieces of this timeline in our earlier look at the shifting AI compliance deadlines facing businesses, and the disclosure obligation is one of the pieces that did not slip.
The United States is arriving from a different direction: state by state. A 2026 AI chatbot compliance overview notes that California's companion-chatbot law, SB 243, took effect January 1, 2026, and requires clear disclosure when a user is interacting with an AI companion that simulates human relationships or anthropomorphic features. A July 13 analysis in Forbes framed the broader wave of chatbot-disclosure laws now taking hold. Our take: when China, the EU, and US state legislatures independently converge on the same rule within a single calendar year, that is no longer a regional quirk. It is the emerging global baseline, and businesses should plan for it as one.
The Line Regulators Are Drawing: Emotional AI Versus Functional AI
The most important detail for most companies is where the line falls, and it is consistent across these regimes. Regulators are not trying to ban AI that helps people. They are trying to constrain AI that pretends to be a person and cultivates an emotional bond.
Companion and persona AI sits on the regulated side. Systems built to simulate friendship, romance, or a persistent human-like relationship are the direct targets. China's measures name this category explicitly. California's SB 243 names it too. This is where security assessments, minor protections, and the heaviest obligations concentrate.
Functional AI sits largely on the exempt or lighter-touch side. Customer service bots, internal workplace assistants, knowledge-base search, and task automation are carved out of China's strictest requirements and, under the EU approach, generally need to satisfy a disclosure obligation rather than a redesign. The systems most businesses actually deploy fall into this category. If you have followed our guidance on building AI customer service that goes beyond the basic chatbot, the good news is that a well-built support assistant that answers questions and resolves tickets is not the thing regulators are trying to shut down.
What this means for businesses: the regulatory risk is concentrated in a specific design choice, not in using AI at all. The moment a functional assistant is dressed up with a fabricated human identity, a fake backstory, or language engineered to make users feel they are bonding with a person, it starts drifting toward the regulated category. The safest and, not coincidentally, the most trustworthy design is an assistant that is genuinely useful and honest about being software.
What This Means for Your Business
Most companies reading this are deploying functional AI, which means the immediate compliance panic is misplaced. But the direction of travel should shape how you design and govern these systems now, because retrofitting disclosure and audit trails after launch is far more expensive than building them in.
First, treat disclosure as a design default rather than a legal afterthought. An assistant that clearly identifies itself as AI at the start of an interaction satisfies the spirit of every regime discussed here and rarely hurts the user experience. Companies that navigate this period well are the ones treating disclosure and persona design as deliberate parts of how they plan and govern AI deployments, rather than decisions left to whoever writes the chatbot's greeting.
Second, keep the persona honest. The temptation to give a support bot a warm human name and an invented personality is strong because it tests well in the short term. The regulatory trend, and frankly the trust trend, is moving against synthetic intimacy in commercial AI. A capable assistant does not need a fake human identity to be pleasant.
Third, know where your systems actually fall. The exemptions are real but conditional. A workplace assistant that quietly evolves into an always-available emotional confidant for employees is not obviously exempt anymore. Map each customer-facing and employee-facing system against the emotional-versus-functional line, and revisit that map as you add features.
A necessary caveat: none of this is legal advice, and the specifics depend on your jurisdiction, your user base, and how your systems behave in practice. The point here is strategic. The terrain your AI runs on is being repaved, and the grading is remarkably consistent across three continents.
How to Prepare Without Overreacting
Practical steps beat both panic and denial.
- Inventory your customer-facing and employee-facing AI. List every system that interacts with a human in natural language, and note whether it is functional, persona-driven, or drifting between the two.
- Add clear AI disclosure by default. Make it unambiguous, at the start of an interaction, that the user is talking to AI. This is cheap to implement and aligns with every regime discussed here.
- Strip out deceptive anthropomorphism. Remove fabricated human identities, invented personal histories, and manipulative emotional framing from functional assistants. Keep tone warm and keep honesty intact.
- Build the audit trail now. Log how your systems behave, what they disclose, and how they handle sensitive interactions. This discipline is the backbone of a workable practical AI governance framework, and it is far easier to add before a system is in production than after.
- Confirm specifics with counsel. Where a system sits near the emotional-interaction line, or where you operate across the EU, US states, or other markets, get qualified legal review rather than guessing.
Common Mistakes to Avoid
The first mistake is assuming this is a China story that stops at China's border. The EU deadline is weeks away and US state laws are already live; the pattern is global. The second mistake is the opposite overcorrection, ripping out useful AI features in a panic when the functional tools most businesses run are largely exempt or only need clear disclosure. The third mistake is confusing a warm user experience with a deceptive one. Regulators are not targeting friendliness; they are targeting AI that hides what it is. A helpful assistant that says plainly it is AI is on the right side of every rule discussed here.
Key Takeaways
- China's Interim Measures for Anthropomorphic AI Interaction Services took effect July 15, 2026, targeting AI that simulates sustained emotional human relationships.
- Functional tools such as customer service bots, workplace assistants, and knowledge search are explicitly exempt from the strictest requirements.
- The EU AI Act's Article 50 chatbot-disclosure obligation applies from August 2, 2026, and US states like California already require companion-AI disclosure, forming a global pattern.
- The line regulators are drawing is between emotional or companion AI, which is constrained, and functional AI, which mainly needs honest disclosure.
- The practical response is to disclose AI by default, avoid deceptive human personas, build audit trails, and confirm specifics with counsel, not to abandon useful AI.
Navigating AI disclosure and anthropomorphism rules does not have to be a solo effort. Book a free discovery call and let's map out what this shift means for how you design and deploy customer-facing AI.