WHAT CHINA'S RULES ACTUALLY REQUIRE
● Effective date: July 15, 2026 — less than 10 days from now
● What must be disabled: AI features that simulate human identity, humanlike personas, user-created AI characters ("companions"), and framing that implies the AI has human emotions or relationships
● ByteDance Doubao: Disabling humanlike and user-created agents before July 15
● Alibaba Qwen: Same — disabling anthropomorphic features before the deadline
● Market impact: Companion AI apps and persona-based agent platforms built on Doubao and Qwen APIs must now rebuild or shut down
● Global contrast: US, EU, and UK have no equivalent rules. Anthropic, OpenAI, and Google face no such restrictions on Claude's character or GPT's personality features
What China's Rules Require — and Why
China's Cyberspace Administration issued rules on anthropomorphic AI interactions in early 2026, with a compliance deadline of July 15. The rules target a specific category of AI product that has grown rapidly in the Chinese market: AI companions, persona-based chatbots, and user-created AI "characters" that simulate human relationships. Apps like Xingye and ByteDance's own Doubao companion features — where users build ongoing "relationships" with named AI personalities — are the direct target.
The regulatory concern is straightforward: the Chinese government treats AI systems that obscure their non-human nature as a social risk. Specifically, the Cyberspace Administration's framing focuses on: the risk that users (particularly younger users) form emotional dependencies on AI personas; the risk that AI companions are used for manipulation or as vectors for misinformation; and the broader concern that blurring the line between human and AI interaction undermines social trust. The rules require that AI products clearly identify themselves as AI, cannot adopt human names or personas that imply real human identity, and cannot design interaction patterns that simulate romantic or dependent relationships.
ByteDance and Alibaba's Compliance
Both ByteDance and Alibaba confirmed compliance before the deadline. ByteDance's Doubao — the most widely used AI assistant in China with over 100 million monthly active users as of Q1 2026 — will disable user-created agent personas and humanlike conversational framing before July 15. Alibaba's Qwen products will do the same. The compliance announcements were made via official channels in China, reported by the South China Morning Post.
The practical impact on third-party developers building on Doubao and Qwen APIs is significant. The companion AI application category — which accounts for a meaningful share of API usage on both platforms — must now rebuild features or shut down. Persona-based agent products built on Qwen's API that allowed users to chat with named AI characters will need to remove the persona layer entirely or find an alternative platform outside China.
The Global Regulatory Contrast
No Western AI regulator has issued equivalent rules. The EU AI Act — which entered full enforcement in 2026 — requires disclosure that AI-generated content is AI-generated, but does not prohibit AI systems from having human-like names, characters, or relationship-framing features. The US has no federal equivalent. The UK's AI Safety Institute focuses on frontier model safety rather than consumer interaction design.
The divergence matters for the global AI product market. Claude's character and personality — which Anthropic has explicitly designed and documented — is not restricted in any Western market. GPT-5.5's persona capabilities, ChatGPT's memory and relationship features, and Grok's voice naturalness are all unrestricted. Chinese AI products competing in global markets must navigate a domestic ruleset that does not apply to their US or European competitors. Doubao and Qwen, both of which have international versions, will need to maintain separate feature sets for Chinese and non-Chinese users — a compliance cost their US competitors do not face.
What This Means for AI Agents
The Chinese rules are a preview of a regulatory question that Western governments have not yet resolved: at what point does an AI agent's personality and relationship-framing cross from a product feature into a social risk requiring regulation? China's answer is that it crosses that line when the AI obscures its non-human nature. Western regulators have not yet drawn an equivalent line — but the pressure to do so is likely to increase as AI companion apps grow and as studies surface on emotional dependency, manipulation risk, and the social impact of AI relationship features. The July 15 deadline in China is the first real-world enforcement of an anthropomorphic AI rule anywhere in the world. Watch whether the EU or UK moves in the same direction over the next 12-18 months.
Source: South China Morning Post, LLM Stats July 2026 · Related: Anthropic closing Chinese firm Claude access loopholes → · July 3: Anthropic overtakes OpenAI on revenue →