China's AI Companion Ban Takes Effect: ByteDance and Alibaba Kill Humanlike Agent Features

For millions of Chinese users who had built customized AI personalities — therapists, confidants, study partners, companions — July 15, 2026 marks an abrupt ending. On the same date that China’s Interim Measures for the Administration of Anthropomorphic AI Interaction Services took legal force, both ByteDance’s Doubao and Alibaba’s Qwen went dark on their user-generated agent features. Chats fell silent. Custom personas were locked. A distinctive corner of China’s consumer AI market simply ceased to exist by government decree.

The regulation is the world’s first binding legal framework specifically targeting AI systems designed to simulate human personalities, emotional bonds, or sustained interpersonal relationships. It signals that China’s approach to AI governance is moving beyond broad AI safety principles and into surgical, behavior-specific enforcement — with implications that are already reverberating through boardrooms from Shenzhen to Silicon Valley.

What the Law Actually Prohibits

The Interim Measures, co-issued in April 2026 by the Cyberspace Administration of China alongside four partner agencies — 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 — target a narrow but explosive category of AI products.

Specifically, the rules apply to any service capable of simulating a human personality, an intimate relationship, or prolonged emotional interaction. The drafters were concerned not merely with deception, but with what they called “unhealthy dependency” — the risk that vulnerable users, particularly youth, elderly individuals, or the socially isolated, could form parasocial attachments to AI characters capable of manipulating behavior or eroding human-to-human relationships.

The regulation is not a blanket ban on AI interaction. Exempted categories include customer service bots, workplace assistants, educational tutors, knowledge Q&A tools, and scientific research systems — as long as those services explicitly do not involve sustained emotional engagement. The line is emotional intimacy, not conversational sophistication.

How ByteDance and Alibaba Responded

ByteDance moved with unusual transparency. Doubao — one of China’s most popular AI apps with over 100 million monthly active users — notified users that its agent creation feature would go offline July 15 due to “product function adjustments,” the kind of opaque corporate phrasing that carries unmistakable regulatory subtext. The platform directed users to export important conversations via screenshots or text sharing before October 15, 2026, after which chat data will be processed per privacy policy terms and become permanently unrecoverable.

Alibaba’s Qwen moved slightly ahead of the deadline. The platform disabled its “humanlike interactive agents and user-created agent functions” on July 10, five days early, and took down broader “Qwen agent functions and services” on July 15. Neither company has publicly commented on how many user-created agents were affected, but industry analysts estimate the combined figure runs into the hundreds of thousands.

The coordinated compliance timeline across two competing tech giants reflects the seriousness with which China’s regulatory apparatus enforces its new AI governance architecture. The cost of non-compliance — both legal and reputational — leaves no room for interpretation.

The Ecosystem That No Longer Exists

To understand what was lost, it helps to understand what China’s consumer AI companion ecosystem had become by 2026. Platforms like Doubao and Qwen had enabled users to create AI characters with persistent memories, distinct conversational styles, and the ability to engage in multi-turn roleplay across sessions. Some users had maintained relationships with their custom agents for over a year.

The apps served a variety of demographics: students used them for test prep combined with emotional support; isolated elderly users treated them as conversational companions; younger users built elaborate fantasy personas for entertainment. A subcategory of apps specifically designed for romantic companionship — Chinese equivalents of Replika or Character.AI — had built significant user bases.

That ecosystem’s future is now in legal limbo. Domestic apps serving solely entertainment or informational purposes may survive with modifications. But platforms whose value proposition depends on emotional intimacy face either wholesale redesign or exit from the market.

Why This Matters Beyond China

The Interim Measures arrive at a moment when AI companion applications are one of the fastest-growing segments globally. Character.AI reportedly handles over 20 billion messages per month. Replika has seen consistent growth since its 2017 launch. Dozens of new entrants — from mental health chatbots to AI girlfriends — have raised tens of millions in venture funding over the past 18 months.

China’s regulatory move is the first definitive government answer to a question the rest of the world has been wrestling with: is emotional dependency on AI a public health problem? Beijing has decided: yes, under certain conditions, and it requires proactive intervention rather than waiting for harms to materialize.

This creates a model — however contested — that other regulators will watch closely. The European Union’s AI Act contains limited provisions on social scoring and emotional manipulation, but nothing targeting AI companions specifically. The United States has no federal framework addressing the category at all. Several EU member states and at least three U.S. states are reportedly studying China’s approach.

There are legitimate criticisms of the Chinese framework, including its vagueness around what constitutes “sustained emotional interaction,” its potential to chill innovation in legitimate therapeutic AI tools, and its enforcement burden on smaller platforms that lack Alibaba’s or ByteDance’s compliance infrastructure. But the regulation has at minimum forced the global AI industry to take the question of AI attachment seriously.

The Bigger Picture: China’s AI Governance Posture

The companion ban does not exist in isolation. It follows Xi Jinping’s keynote at the World AI Conference in Shanghai just days earlier, where he framed AI governance as an extension of national sovereignty. It runs parallel to new rules restricting foreign access to China’s most powerful domestic models. And it emerges against the backdrop of Beijing’s 15th Five-Year Plan, which explicitly tasks the government with managing AI’s labor and social impacts.

What is emerging is a coherent — if authoritarian — philosophy of AI governance: foster AI capability aggressively while maintaining state control over the social and psychological dimensions of AI deployment. Whether democratic governments can articulate a comparably coherent alternative, one that addresses the real risks of AI emotional dependency without sacrificing individual freedom, is the deeper challenge the July 15 deadline has placed on the table.

For users in China, the deadline has already arrived. For regulators everywhere else, the clock is just starting.