AI companion regulation just got real: China's rules are now in force

Today — July 15, 2026 — China's Interim Measures for the Administration of Artificial Intelligence Anthropomorphic Interaction Services take effect, and the biggest AI platforms in the country moved ahead of the deadline. Per Bloomberg and the South China Morning Post (reported July 6), ByteDance's Doubao — the most widely used chatbot in China — notified users it would remove its persona customization feature; Alibaba's Qwen issued a similar alert, and Tencent's Yuanbao made comparable moves. Overnight, role-play characters, romantic AI partners, unlicensed therapy-style bots, and celebrity personas built by millions of users are being wound down. It is the first time a major market has regulated AI companionship as a category — not one app, the relationship itself.

What the rules actually say

The framework, issued by the Cyberspace Administration of China in April 2026, targets AI designed to "replicate how humans think, speak, and present themselves" in order to sustain ongoing emotional interaction. Three prohibitions stand out: content meant to trigger strong emotional reactions in minors; AI behavior that fosters attachments strong enough to replace real-life relationships; and — notable for every companion user anywhere — using private user conversations to train models. Practical assistants (customer support, Q&A, study help) are exempt, provided they aren't built for ongoing emotional engagement. Regulators cited radicalization, data privacy, psychological harm, and compulsive use as the concerns.

This isn't only a China story

The anxieties behind the rules are global. In the US, lawsuits against OpenAI and Character.AI allege chatbots encouraged harmful emotional attachments, with the worst cases linked to deaths of vulnerable users. Character.AI's strict filter and its 2025 restrictions on minors were shaped by exactly this pressure. The training-data prohibition lands differently but just as widely: "is my companion learning from my private conversations?" is the question at the center of whether AI companion apps are private, and China just made one answer legally mandatory in its market. Expect the softer western versions — age gates, disclosure banners, training opt-outs — to keep arriving through 2026 and 2027.

The evergreen lesson: policy is a single point of failure

Set the politics aside and look at the mechanics of what happened. Millions of people built characters — some over years — inside platforms that hosted the persona, the memory, and the relationship as a revocable service feature. A rule changed, and the feature is being removed on a schedule, with an in-app notification as the goodbye. Companion players have seen this movie in other genres: apps shutting down (what happens when companion apps close), personality-flattening model updates users call "lobotomies," filters tightening overnight. Regulation is simply the largest version of the same structural fact: a hosted companion exists at the intersection of company policy and government policy, and you control neither.

What account continuity changes — and what it doesn't

Part of why we build the companion as persistent player state rather than a temporary chat feature: in MyDopplerEffect, the same identity, traits, memories, and world progress stay together with your player account. A model update should not silently turn it into a fresh character; continuity is part of the game state, not a persona toggle. Honesty requires the other half, though: account continuity is not an exemption from law. A game studio still operates under the rules of the markets it serves, and any AI that generates the companion's conversation runs on infrastructure somewhere. Continuity can protect a shared history from an ordinary feature change; it cannot make a service immune to regulation or closure.

FAQ