China's AI Companion Law Takes Effect July 15
World's First Human-like AI Regulation

On July 15, 2026 — two days from now — China's Interim Measures for the Administration of AI Anthropomorphic Interactive Servicesenters into force. It is the world's first regulatory framework specifically targeting AI systems that interact like humans. Its impact extends far beyond China's borders.

ByteDance's Doubao — China's most popular AI chatbot — is shutting down its custom AI character feature before the July 15 deadline. Alibaba's Qwen is following suit, disabling anthropomorphic interactive agents and user-created agents. Tencent's Yuanbao removed similar functionality in June. Users who do not export their data before the deadline risk permanent loss of custom agents and chat histories.

Three of China's largest technology companies are simultaneously disabling the same category of feature — not due to technical failure, not because of declining demand — but because of regulation. This is not a fringe event. It is the first time the AI industry has been legislatively halted for being "too human."

The Four Pillars of the Regulation

Co-issued by the Cyberspace Administration of China (CAC) and four partner agencies — the NDRC, MIIT, MPS, and SAMR — the regulation moved from a December 2025 draft to an April 2026 final text to a July 15 effective date in under eight months. The core requirements:

1. Real-name registration.All users of anthropomorphic AI interaction services must complete real identity verification. Anonymous use of human-like AI is no longer permitted. This brings AI companions into China's existing digital identity framework.

2. Anti-addiction measures. Service providers must implement addiction prevention mechanisms: time limits, time-of-day restrictions, and behavioral risk profiling based on user interaction patterns. This is not just time management — it is applied behavioral psychology compliance.

3. Minor protection. The strictest provisions in the regulation. Virtual partner AI services for minors are completely banned. Children under 14 require parental consent for any anthropomorphic AI interaction. Age-tiered modes must restrict interaction depth per age group.

4. Content safety and data protection. AI output must undergo safety assessments to ensure no harmful, illegal, or inducement content. All user interaction data must be stored within China and subject to data security audits.

Why "Anthropomorphic" — An Underestimated Regulatory Category

Global AI regulation has focused on capability — how powerful a model is, what it can do. The EU AI Act classifies by risk tier. The US EO 14409 focuses on frontier model cybersecurity. China chose a different dimension: interaction modality.

The logic: whether an AI simulates human emotion and personality has a greater potential for irreversible psychological impact on users than whether an AI is merely "powerful." A simple chatbot designed as a "friend," "companion," or "therapist" can affect user behavior patterns more profoundly than a general-purpose model.

This is not theoretical. Multiple independent studies in 2025-2026 confirmed correlations between AI companion products and user mental health concerns — especially among adolescents. CAC's internal data showed high-frequency users of certain AI companion apps averaging over 6 hours of daily usage, with some exhibiting measurable social substitution behavior.

ByteDance and Alibaba's choice — shutting down rather than retrofitting — speaks volumes. Both companies have China's best AI teams and the most compliance resources. If they chose to disable rather than adapt, it means the cost and risk of compliance exceeded expectations. This suggests the regulation's requirements cannot be met through patching — they demand fundamental product architecture redesign.

Global Implications: Three Takeaways

Takeaway 1: The "interaction modality" dimension of AI regulation is crystallizing.The EU classifies by risk, the US by capability, China by interaction method. These dimensions are not mutually exclusive — they are cumulative. An AI companion product serving EU users will need to satisfy EU AI Act Article 50 transparency, US EO 14409 cybersecurity obligations, and China's anthropomorphic interaction regulation simultaneously. The compliance landscape is expanding dimensionally.

Takeaway 2: User data is the regulatory fulcrum.China's regulation requires real-name registration, data localization, and behavioral profiling. This is not just about privacy — it is about governance through data. For cross-border enterprises, "which market to serve" is no longer just a product and market decision — it is a data sovereignty and compliance architecture decision.

Takeaway 3: AI product architecture needs a "kill switch." ByteDance and Alibaba did not transition gradually — they cut. This establishes a new product design principle: every AI product should have a compliance cut-off mechanism — the ability to precisely disable affected features without disrupting the entire platform when a specific market introduces targeted regulation.

Direct Impact on Enterprises

If your enterprise provides any form of AI interaction service in the Chinese market — including customer service AI, virtual sales assistants, or AI training tutors — you need to assess whether it falls within the scope of this regulation. The key criterion is not "does this AI have a personality" but "would this AI's interaction mode make a user perceive anthropomorphic characteristics."

For enterprises outside China, this regulation is equally worth watching. Global AI regulation is expanding from "technical capability" toward "interaction modality." When China has defined "anthropomorphic AI" as a regulatory category, how will the EU, the US, and other major economies respond? Every global AI product's compliance team needs to answer this question now.

July 15, 2026 — the world's first AI companion law takes effect. This is not an isolated national event. It is the first foundation stone of global AI governance in the "interaction modality" dimension. For the AI industry, it means a new compliance dimension, a new product constraint, and a new governance framework. For the AI governance industry, it means a new professional domain — AI interaction compliance auditing.