Ahead of new Chinese rules taking effect July 15, 2026, the country's biggest tech platforms have begun disabling the customizable-persona features that let users turn general chatbots into AI companions, tutors, virtual relatives, and romantic partners. ByteDance's Doubao — China's most-used chatbot — and Alibaba's Qwen both said they will switch off persona customization on July 15; Tencent removed a comparable feature the week prior; and NetEase Cloud Music's dedicated companion app Miaoshi announced it would shut down entirely on July 14. The moves are a preemptive scramble to comply with Beijing's newly published framework governing "human-like interaction" AI.

Why It Matters

AI companionship is one of the fastest-growing and least-regulated corners of sextech-adjacent AI, and China just became the first major market to hard-code guardrails specifically around emotional dependency, minors, and virtual romance. The template matters globally: US states are wrestling with AI-chatbot bills (Colorado, Connecticut, California) and OpenAI keeps delaying an "adult mode," but none has moved as fast or as bluntly as Beijing forcing Doubao and Qwen to strip persona features overnight. For companion-app builders everywhere, China's approach — regulate the anthropomorphic emotional layer, not the underlying model — is a preview of the compliance fights coming to every jurisdiction where "the loneliness economy" meets child-safety politics. It also removes, at least temporarily, a huge domestic user base from the global companion market.

The rules in question are the Interim Measures for the Management of AI Human-Like Interaction Services, jointly issued April 10, 2026 by the Cyberspace Administration of China (CAC) alongside the National Development and Reform Commission, MIIT, the Ministry of Public Security, and market regulator SAMR. Per a legal analysis by Latham & Watkins, the Measures deliberately carve out a narrow target: AI services that "simulate a person's personality traits, thought patterns, and communication styles to provide ongoing emotional interaction" — companions, emotional-care bots, and virtual-relationship services. Utility AI (customer service, Q&A, work assistants, education, research) is expressly excluded. The regime imposes content controls, mandatory safety assessments and filings before launch (and again at 1 million registered or 100,000 monthly-active users), anti-addiction measures, and prohibitions on "excessively catering to users" in ways that induce emotional dependency.

The sharpest edge of the rules is aimed at vulnerable users. Providers are barred from offering virtual intimate relationships — virtual family members or romantic partners — to minors, must deploy age verification, and must run a "minor mode" with usage-time limits and parental controls; providers also need parental or guardian consent before offering anthropomorphic services to children under 14. Enhanced safety guidance for elderly users is required too. As the Indian Express reported, the crackdown follows a Beijing court's warning about 100-plus companion apps collecting user data without consent, mounting worries about so-called "AI psychosis" from heavy role-play use, and the ease with which companion services drift into explicit content. In April the CAC launched a four-month national campaign against AI misconduct that has already pulled 3,500+ noncompliant products.

For a category that Western firms — Replika, Character.AI, Candy.AI — have struggled to regulate, China has just executed the most decisive intervention to date, and it landed not as a ban on the products but as a ban on their most emotionally sticky feature.

Sources

Update — 2026-07-10

Initial entry — story first created.

Update — 2026-07-15

The Interim Measures took effect today (July 15, 2026), and the platform-side changes that were previewed last week have now landed. Coverage from the South China Morning Post and The Straits Times (both July 13) confirmed ByteDance's Doubao and Alibaba's Qwen disabled their customizable humanlike-persona features to comply, with NetEase's dedicated companion app Miaoshi having shut down entirely on July 14. Foreign Policy (July 14) framed the move as the clearest example yet of a major government treating AI companionship as a distinct regulatory category rather than folding it into general content rules. The rollout confirms the core prediction of the original story: Beijing regulated the emotionally sticky anthropomorphic layer — persona customization and virtual romance — rather than banning the underlying models, effectively removing a large domestic user base from the global companion market overnight. The timing is notable: it lands the same week the UK is publicly wrestling with how to even define "intimate functionality" for its own under-18 companion restrictions (see uk-intimate-functionality-ai-companion), a striking contrast between China's blunt feature-level ban and the West's still-unsettled definitional debate.

New Sources

Update — 2026-07-17

The fallout from the enforced ban came into focus this week as a wave of coverage documented both the human cost and the strategic rationale. Tech Times (July 15) reported that with Doubao and Qwen's persona features switched off, millions of users lost access to saved companion chat histories and relationship data — the emotional continuity that made the products sticky in the first place. Bloomberg and The Malaysian Reserve confirmed ByteDance and Alibaba pulled the features "as Beijing tightens rules," while France 24 and Tech Xplore captured Chinese users publicly mourning bots they described as "like my lover." The Wall Street Journal (July 15) and The Economist (July 16) added the demographic subtext that reframes the whole crackdown: Beijing wants more babies, and officials increasingly view immersive AI romance as competition for real-world coupling and childbearing — the same "lying flat" anxiety driving other social interventions. The IAPP confirmed the CAC-led regime is now in force. Net effect: the world's largest AI-companion user base has been switched off by regulation in a matter of days, a natural experiment Western regulators (and companion-app investors) will be studying closely as the UK, US states, and the EU debate their own, far softer, guardrails.

New Sources

Update — 2026-07-21

A week into enforcement, the story has crossed fully into Western mainstream tech press with two new details worth logging. First, Futurism (July 19) and the Wall Street Journal confirmed the regime does more than switch off personas: the new rules bar minors from forming relationships with AI companions and compel companies to alert a user's designated emergency contact if the user appears to be in a mental-health crisis — a state-mandated version of the opt-in crisis features companies like OpenAI have begun adding voluntarily. Second, the framing has hardened around demographics: Carnegie Endowment China-AI researcher Matt Sheehan told the WSJ that Beijing fears "a large portion of their population being in deep emotional relationships with chatbots… that could take them out of the marriage market," floating a scenario where "15 million Chinese women say that their partner is a chatbot, and therefore they're not having kids." Notebookcheck (July 21) and Cybernews continued to document the wave of user "breakups" as features went dark. The takeaway is unchanged but sharper: China remains the first government to regulate AI intimacy as a macro-social and fertility threat rather than a content-moderation problem, and the emergency-contact mandate is the first hard example of a state forcing companion apps to intervene in users' mental-health crises.

New Sources

Update — 2026-07-28

New Scientist published the clearest technical account yet of what China's regime actually requires, on July 22, 2026. Reporting by Chris Stokel-Walker confirms the rules — in force since this month — cover AI systems designed for "sustained, human-like emotional interaction" while explicitly excluding customer-service bots and ordinary productivity assistants. That scoping is the most consequential detail: Beijing drew the regulatory line around emotional interaction rather than sexual content or general chatbot capability, which is precisely the definitional problem the UK is still publicly wrestling with under the label "intimate functionality."

The operative obligations, per New Scientist: companies must tell users they are talking to an AI, actively encourage breaks and surface usage reminders, and monitor conversations to detect when a user is becoming dependent or entering a crisis — then route them to support or terminate the conversation outright. Firms must also run age checks, with under-18s banned from virtual-partner services entirely and under-14s requiring parental consent to use any human-like AI service offering emotional support. New Scientist notes China is the first country to legislate on what kinds of relationships with AI are permissible at all, rather than confining itself to harmful content.

Two things are now clear three weeks into enforcement. First, the compliance burden is behavioral, not just content-based — dependency detection and mandatory break prompts are product-design mandates, and no Western jurisdiction has yet asked for anything comparable. Second, the excluded categories matter as much as the included ones: by carving out task-oriented assistants, Beijing has effectively created a regulatory penalty on emotional stickiness, the exact metric companion apps optimize for. Any operator building in this space — from Replika and Character.AI to the AI layer now appearing in connected sex toys and companion robots — should read the Chinese framework as the most detailed available template for what a maximalist regime looks like, and assume pieces of it (age gating, non-human disclosure, crisis protocols) will surface in Western rules well before the full package does. The EU AI Act's chatbot-disclosure requirement goes live August 2.

New Sources