China's AI Companion Reckoning: Doubao and Qwen Shut Down Agent Features as New Rules Take Effect July 15

`json { "skip_run": false, "title": "China's AI Companion Reckoning: Doubao and Qwen Shut Down Agent Features as New Rules Take Effect July 15", "excerpt": "ByteDance's Doubao and Alibaba's Qwen are pulling their custom AI agent features on July 15, 2026, as China's landmark Interim Measures for the Administration of AI Anthropomorphic Interactive Services come into force — a regulatory reset that is reshaping how the world's largest AI market thinks about emotional AI, user safety, and the line between companion and tool.", "tags": ["China AI", "ByteDance", "Doubao", "Alibaba", "Qwen", "AI Regulation", "CAC", "AI Companions", "Tencent", "Anthropomorphic AI", "AI Policy", "Developer Tools"], "sources": [ {"title": "ByteDance's Doubao and Alibaba's Qwen to Shut Down AI Agent Features on July 15 — TechNode", "url": "technode.com↗"}, {"title": "China AI Companion Law Arrives July 15: Doubao, Qwen Agent Data Will Be Deleted — TechTimes", "url": "techtimes.com↗"}, {"title": "ByteDance and Alibaba Disable Humanlike AI Custom Agents as New Rules Loom — South China Morning Post", "url": "scmp.com↗"}, {"title": "China's AI Companion Rules Force Doubao, Qwen Shutdowns — ChinaTechNews", "url": "chinatechnews.com↗"}, {"title": "China's New Regulations on AI Anthropomorphic Interactive Services — Two Birds Law", "url": "twobirds.com↗'s-new-regulations-on-ai-anthropomorphic-interactive-services"}, {"title": "China Humanlike AI Agent Rules — The Next Web", "url": "thenextweb.com↗"}, {"title": "China's AI Companions Vanish Ahead of New Rules — Artiverse", "url": "artiverse.ca↗"}, {"title": "China Bans AI Partners for Minors — AI Safety China Substack", "url": "aisafetychina.substack.com↗"}, {"title": "China Issues First National Policy Framework Dedicated to AI Agents — NYU Shanghai RITS", "url": "rits.shanghai.nyu.edu↗"}, {"title": "Doubao's Paid Pivot Signals ByteDance's AI Monetization Push — China Biz Insider", "url": "chinabizinsider.com↗"}, {"title": "China Writes the AI Companion GDPR for a $30B Market — Forbes", "url": "forbes.com↗"}, {"title": "China's Interim Measures for AI Anthropomorphic Interactive Services — Geopolitechs", "url": "geopolitechs.org↗"}, {"title": "Rimon Law China AI Law Brief — July 2026", "url": "rimonlaw.com↗"}, {"title": "Global Times: China AI Companion Regulation July 2026", "url": "globaltimes.cn↗"}, {"title": "Qwen3.7 The Agent Frontier — Alibaba Cloud Blog", "url": "alibabacloud.com↗"}, {"title": "China AI Companion Market Outlook to 2030 — Grand View Research", "url": "grandviewresearch.com↗"} ] } ```

China's AI Companion Reckoning: Doubao and Qwen Shut Down Agent Features as New Rules Take Effect July 15

On July 15, 2026, something quietly significant happened in the world's largest AI market: ByteDance shut down the custom agent creation feature inside Doubao, its flagship AI app with roughly 330 million monthly active users, and Alibaba disabled the humanlike interactive agents inside Qwen, which counts 234 million users on its mobile platform alone. Neither company framed the move as a retreat. Both described it as compliance. But the underlying story is more consequential than a routine regulatory checkbox — it marks the moment China drew a formal, enforceable line between AI as a productivity tool and AI as an emotional companion.

The trigger is the *Interim Measures for the Administration of AI Anthropomorphic Interactive Services*, jointly issued on April 10, 2026, by five national authorities — the Cyberspace Administration of China (CAC), the National Development and Reform Commission, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the State Administration for Market Regulation. The measures took effect on July 15, 2026, and they represent China's first dedicated regulatory framework↗ for AI services that simulate human personality, thinking patterns, and communication styles to provide "continuous emotional interaction."

For global developers and enterprise buyers watching China's AI ecosystem, this is not a niche compliance story. It is a signal about where the world's most prolific AI market is heading — and what it means for the products, architectures, and business models that will survive.

What the Law Actually Says

The *Interim Measures* are precise about their scope. They apply to AI services that provide "continuous emotional interaction" by simulating human traits — virtual companions, digital personas, emotional chatbots, and user-created characters with persistent memory and fixed speaking styles. Standard productivity tools, customer service bots, knowledge Q&A assistants, and workplace AI are explicitly excluded, provided they do not facilitate sustained emotional bonding.

The compliance requirements are technically demanding, and that is precisely why Doubao and Qwen chose shutdown over retrofit:

• Anti-addiction friction: Platforms must issue mandatory pop-up notifications after two continuous hours of interaction, and must provide an "instant-exit" mechanism that stops the service immediately upon user request — no manipulative retention tactics permitted.
• Transparency mandates: Services must clearly disclose their artificial nature to users at all times, with dynamic reminders triggered when signs of over-dependence are detected.
• Minor protection: Providers are strictly prohibited from offering virtual intimate relationship services — virtual partners, virtual relatives — to minors. Services for children under 14 require explicit parental consent, and a dedicated "minor mode" with content filtering, usage time limits, and guardian controls is mandatory.
• Crisis intervention: When a user exhibits signs of self-harm or extreme psychological distress, the system must provide supportive content, enable manual operator escalation, and notify emergency contacts or guardians.
• Data governance: Interaction data cannot be used for model training without separate, explicit user consent. Users must be granted the right to delete their chat history. Encryption and access controls are mandatory.
• Regulatory filing: Services reaching 1 million registered users or 100,000 monthly active users must complete algorithm registration and undergo mandatory security assessments with provincial cyberspace authorities.

Fines for violations range from RMB 10,000 to RMB 200,000 for cases resulting in actual harm to citizens' physical or mental health, with the possibility of service suspension for repeat or severe offenders.

"The regulation adopts an 'inclusive and prudent' but 'classified and graded' oversight model — balancing encouragement of technological innovation with strict requirements for national security and public interest," notes the Two Birds law firm analysis↗ of the measures.

Why Doubao and Qwen Chose Shutdown Over Retrofit

The architectural incompatibility is the crux of the story. Persistent-memory AI agents — the kind that remember your name, your preferences, your emotional history, and build a sense of continuity across sessions — are designed to minimize friction and maximize engagement. The *Interim Measures* mandate the opposite: mandatory friction, mandatory interruptions, mandatory exits. Retrofitting a companion agent to feel like a compliance tool while still functioning as a companion is, in practice, a contradiction.

ByteDance announced that Doubao's agent creation feature would go offline on July 15, 2026. Users were granted a read-only grace period until October 15, 2026, to view and export their agent configurations and conversation histories — after which the data will be permanently deleted. ByteDance has directed users toward its Maoxiang application as the designated migration path for future agent creation, signalling that the company intends to continue in the space, just under a compliant architecture.

Alibaba followed a tighter timeline. According to TechNode's reporting↗, Qwen began disabling humanlike interactive agents and user-created agent functions as early as July 10, with full shutdown on July 15. Unlike ByteDance, Alibaba did not announce a clear migration path or alternative platform for users, and agent configurations and conversation histories are slated for permanent deletion. For users who had invested months building detailed character configurations, the loss is total.

Tencent moved earlier. Its Yuanbao assistant removed custom agent features in June 2026, ahead of the enforcement deadline — a sign that the largest Chinese tech firms had been reading the regulatory direction clearly since the April promulgation.

The South China Morning Post↗ reported that users expressed frustration over the loss of long-standing "emotional" connections, with some describing agents they had interacted with daily for over a year. The platforms are prioritizing compliance to avoid the legal risks associated with their previous, more open-ended agent architectures.

The Broader Regulatory Architecture

The *Interim Measures* do not stand alone. They are part of a coordinated July 15, 2026, regulatory package that also includes the *Implementation Opinions on the Standardized Application and Innovative Development of Intelligent Agents* — China's first national policy framework dedicated to AI agents↗ as a distinct regulatory category.

The *Implementation Opinions* define AI agents as intelligent systems capable of autonomous perception, memory, decision-making, interaction, and execution — and apply a risk-based tiered oversight model. Agents in sensitive sectors (healthcare, transportation, media, public safety) face mandatory filing, compliance testing, and product recall provisions. The policy sets an ambitious target: 70% adoption of intelligent agents in smart terminals by 2027, with explicit encouragement to integrate with domestic chips and operating systems.

Together, the two frameworks signal a deliberate strategic move: China is not retreating from AI agents. It is bifurcating the market. Emotional companion AI gets regulated into compliance or out of existence. Productivity and enterprise AI gets a policy tailwind and an industrial roadmap.

What This Means for the Companion AI Market

The numbers make the stakes clear. China's AI companion market is projected to reach USD 12.58 billion by 2030, growing at a CAGR of 35.4% from 2025 to 2030, according to Grand View Research↗. The July 15 measures do not kill that market — they reshape it. Services that can demonstrate compliance (age-gating, anti-addiction systems, transparent AI disclosure, crisis intervention) will be permitted to operate. Those that cannot, or choose not to, will exit.

The practical effect is a consolidation toward enterprise-grade, auditable AI infrastructure. As Artiverse's analysis↗ notes, the industry is moving away from "unregulated" consumer-facing companion bots toward tools that prioritize compliance, safety, and commercial viability over raw user engagement.

The Doubao Context: From 60M to 330M Users in 18 Months

To understand the scale of what is being regulated, consider Doubao's trajectory. In November 2024, Doubao had approximately 60 million monthly active users. By April 2026, that number had surged to approximately 336 million, making it the second-largest AI application globally, trailing only ChatGPT. ByteDance introduced a tiered subscription model in June 2026 — Standard, Enhanced, and Premium plans — which resulted in a modest decline to roughly 330 million MAU as casual users churned. For context, Alibaba's Qwen sits at approximately 166 million MAU in the domestic market, and DeepSeek at roughly 127 million MAU, according to China Biz Insider's monetization analysis↗.

The agent features being shut down were a meaningful part of Doubao's engagement flywheel — users who built persistent characters were among the most active and retained. Losing them is a real cost, which is why ByteDance's Maoxiang migration path matters: it is an attempt to retain the user relationship while moving the architecture into compliance.

Practical Takeaways for Developers and Enterprise Buyers

What Is Shutting Down

• Doubao: Custom agent creation, persistent-memory character configurations, and associated conversation histories. Read-only access until October 15, 2026; permanent deletion thereafter. Migration path: Maoxiang app.
• Qwen: Humanlike interactive agents and user-created agent functions, disabled from July 10–15, 2026. No announced migration path. Agent data subject to permanent deletion.
• Tencent Yuanbao: Custom agent features removed in June 2026.

What Is Not Shutting Down

The *Interim Measures* explicitly exclude productivity tools, customer service bots, knowledge Q&A assistants, research tools, and workplace AI — provided they do not facilitate sustained emotional interaction. This means:

• Qwen's model API (including Qwen3.7-Max and the broader Qwen3 family) remains fully available via Alibaba Cloud Model Studio↗ for developers building productivity and enterprise applications.
• Doubao's core AI capabilities — text generation, coding assistance, multimodal features — continue to operate under the subscription model.
• DeepSeek, GLM-5.2, Kimi, and other API-first models are unaffected, as they are positioned as developer infrastructure rather than consumer companion services.

How to Build Compliantly

For developers building on Chinese AI infrastructure who want to include any form of persistent persona or character interaction, the compliance checklist is now clear:

• Implement mandatory two-hour usage notifications and instant-exit mechanisms.
• Disclose AI identity explicitly and dynamically.
• Build age-gating and parental consent flows for users under 14.
• Implement crisis detection and intervention protocols.
• Obtain separate user consent before using interaction data for model training.
• File algorithm registration with the CAC if you exceed 1 million registered users or 100,000 MAU.

The Global Signal

China's *Interim Measures* are the most detailed national regulatory framework for AI companion services anywhere in the world. The EU's AI Act addresses high-risk AI systems broadly; the US has no equivalent federal framework. For global AI companies operating in or selling into China, the July 15 enforcement date is a hard deadline — not a guideline.

More broadly, the framework offers a preview of where other regulators may head. The core concerns — addiction, minor protection, emotional manipulation, data governance — are not unique to China. The specific mechanisms China has chosen (mandatory friction, crisis intervention, algorithm registration) are a policy template that other jurisdictions will study.

For the Chinese AI ecosystem itself, the message is clear: the era of unchecked growth in consumer-facing emotional AI is over. What replaces it is a more structured, more auditable, and ultimately more commercially durable market — one where the winners are the labs and platforms that can build compliant infrastructure at scale. ByteDance, Alibaba, and Tencent have the resources to do that. The question is whether the compliance cost creates an opening for smaller, more agile players — or simply consolidates the market further around the giants who can afford to build twice.

The agents are going offline. The race to build what comes next has already begun.