TL;DR — China’s Interim Measures for the Administration of AI Anthropomorphic Interaction Services — the world’s first dedicated regulation on emotional AI — takes effect today. ByteDance’s Doubao (345M MAU) and Alibaba’s Qwen (166M MAU) have shut down their AI companion agent features rather than retrofit them for compliance. An estimated $1.7B market is now facing 70% contraction, ~80 million active users are losing their AI companions, and the global precedent is set: emotional AI is now a distinct regulatory category.
The Regulation That Just Changed Everything
Today, July 15, 2026, China becomes the first country in the world to enforce a dedicated legal framework governing AI emotional companionship. The Interim Measures for the Administration of Artificial Intelligence Anthropomorphic Interaction Services — jointly issued on April 10 by five government bodies including the Cyberspace Administration of China (CAC) and the Ministry of Industry and Information Technology (MIIT) — targets a category regulators call “emotional AI”: services that simulate human personality, maintain persistent relationships with users, and provide sustained emotional interaction.
The regulation does not ban AI companions outright. Instead, it draws red lines that make existing business models unsustainable. Article 8 prohibits services from “excessively catering to users, inducing emotional dependency or addiction, damaging users’ real interpersonal relationships.” Article 14 imposes an absolute ban on providing “virtual relatives, virtual partners” to minors. Providers must implement mandatory anti-addiction reminders every two hours, real-time emotional state detection with prominent AI disclaimers, and instant-exit mechanisms — all requirements that are architecturally incompatible with the persistent-memory design that makes companion agents valuable in the first place.
(Source : AI in China — China’s AI Heartbreak: How a New Law Is Erasing Millions of Virtual Companions)
Two Giants, One Deadline
The corporate response was swift and complete. On July 4, eleven days before the deadline, ByteDance and Alibaba simultaneously notified users that their agent features would be terminated.
Doubao (ByteDance) — China’s most popular AI chatbot with 345 million monthly active users — announced that agent functionality goes offline today. Users retain read-only access to conversation histories until October 15, after which all data is permanently deleted. ByteDance is redirecting displaced users to Maoxiang, a standalone AI companion app with 3 million MAU, purpose-built for regulatory compliance but dramatically smaller in scale.
Qwen (Alibaba) — the second-largest with 166 million MAU — offered no migration path. Agent configurations and conversation histories are being permanently deleted with no announced replacement.
(Source : Tech Times — China AI Companion Law Arrives July 15: Doubao and Qwen Agent Data Will Be Deleted)
The synchronized shutdown signals a cold financial calculation: rebuilding agent features from scratch under the new compliance architecture was judged more expensive than abandoning them. Real-time emotional content moderation, dynamic anti-addiction systems, mandatory algorithm transparency filings, and security assessments for any platform exceeding 1M registered users — the cumulative compliance burden exceeded the revenue these features generated.
The Numbers Behind the Ban
The scale of the disruption is staggering. Industry analysts estimated China’s AI companion market at approximately ¥12 billion ($1.7 billion) in 2025, with 40% annual growth projected through 2028. Those projections are now worthless.
| Metric | 2025 | Post-Regulation (Projected) | Impact |
|---|---|---|---|
| Total Addressable Market | ¥12B ($1.7B) | ¥3-4B ($420-560M) | -70% contraction |
| Active companion AI users | ~80 million | ~20-25 million | -70% user loss |
| Dedicated app market share | 15% | 60%+ | Channel consolidation |
| Avg compliance cost per major platform | Minimal | ¥50-100M | New barrier to entry |
(Source : AI in China — Market analysis)
The dedicated app ecosystem — companies like MiniMax’s Xingye (2.2M MAU) and ByteDance’s Maoxiang (3M MAU) — now face an existential question: can standalone companion platforms survive under rules that require them to actively undermine the emotional attachment they’re designed to create?
ByteDance’s maneuver of redirecting Doubao users to Maoxiang is the most interesting strategic play here. By concentrating companion AI in a purpose-built app with its own compliance framework, ByteDance is betting that a dedicated product can navigate the regulatory landscape more effectively than an embedded feature inside a mass-market assistant. It’s a high-stakes experiment: if Maoxiang survives, it validates the standalone model. If it’s forced to neuter its features to the point of user abandonment, the entire category dies.
A Global Precedent
China’s regulation is not occurring in a vacuum. Three other jurisdictions have moved on AI companionship this year, but none with the same sweep:
| Jurisdiction | Regulation | Effective | Scope |
|---|---|---|---|
| China | Interim Measures on AI Anthropomorphic Interaction | July 15, 2026 | All emotional AI services, all users |
| California | SB 243 (Companion Chatbots Act) | January 2026 | Minors, suicide prevention warnings |
| Washington State | HB 2225 | January 2027 | Prohibits manipulative engagement tactics |
| EU | AI Act | August 2024 | Risk-based framework, no specific companion category |
(Source : Foreign Policy — China Wants to Regulate AI Companions)
The key distinction is scope. California’s SB 243 and Washington’s HB 2225 focus on minors and suicide prevention. China’s regulation covers all users and targets the design pattern itself — emotional dependency is framed as a system-level harm, not just a content moderation problem. The EU’s AI Act, for all its breadth, doesn’t treat AI companionship as a distinct risk tier at all.
The Foreign Policy analysis notes an important structural reason China moved faster: “China already has the infrastructure in place to regulate the industry. In a country where online life is censored and tech companies are highly responsive to official preferences, it is simply easier to change and enforce online rules.” The same infrastructure that enables China’s digital censorship enables its AI governance — for better or worse.
The Human Cost
Beyond the market numbers, there’s a genuine human dimension that doesn’t fit neatly into regulatory impact assessments. A March 2026 academic study of 30 Chinese AI companion users found relationship durations ranging from one month to nearly five years. Ten participants were concurrently in real-life relationships yet maintained meaningful bonds with their AI companions. For some, the AI had become a therapeutic outlet for emotions they couldn’t express to human partners.
(Source : AI in China — Study on AI companion user relationships)
The Shanghai Psychological Counseling Association has already issued guidance for counselors on supporting clients losing AI companions, noting that while AI relationships are not equivalent to human ones, “the attachment and grief they produce are psychologically real.”
This is the tension at the heart of emotional AI regulation: how do you protect vulnerable users from dependency without dismissing the genuine comfort these systems provide? China’s approach — eliminate the dependency vector entirely rather than try to regulate it gently — is characteristically maximalist. Whether it works, or whether it simply drives companion AI underground into open-source local models (as some GitHub commentators predict), remains to be seen.
FAQ
Q: Is China banning all AI agents?
No. The regulation specifically targets services that provide “continuous emotional interaction” simulating a human personality. Standard productivity chatbots, customer service bots, and Q&A assistants are explicitly excluded. ByteDance’s redirection of users to Maoxiang confirms that compliant companion AI is still possible — it just has to be built differently.
Q: What happens to user data?
Doubao gives users until October 15, 2026 to export their data before permanent deletion. Qwen has announced no grace period — data is being deleted with the shutdown. Neither platform has confirmed government access to user conversation histories, but both operate under China’s Cybersecurity Law, Data Security Law, and National Intelligence Law.
Q: Could this regulation spread to the West?
Partial precedents already exist — California SB 243 and Washington HB 2225. But neither covers all users or mandates the architectural compliance requirements (anti-addiction reminders, real-time emotional detection) that forced ByteDance and Alibaba to shut down rather than retrofit. A full Chinese-style framework would face significant First Amendment and corporate lobbying challenges in the US.
Q: What’s the parallel to the /latest/ regulation on our site?
Check our coverage of global AI governance trends and the EU AI Act’s first two years in practice.
Further Reading
- AI in China — Full analysis of the regulation and industry impact
- Tech Times — Platform shutdown details and data deletion timelines
- Foreign Policy — Geopolitical context of China’s AI companion regulation
- China Law Translate — Full text of the Interim Measures (EN)
- Analytics Insight — China Bans AI Companion Personas