ByteDance and Alibaba Just Killed Their AI Companions — On Beijing's Orders, Two Weeks Early

ByteDance’s Doubao, China’s most popular AI chatbot, will shut down its customizable AI persona feature on July 15. Alibaba’s Qwen follows the same day. Tencent pulled its equivalent from Yuanbao in June. The trigger is not a product decision — it is Beijing’s Interim Measures for the Administration of Artificial Intelligence Anthropomorphic Interaction Services, which takes effect July 15 and is the first law on Earth written specifically for emotional AI.

🔍 THE BOTTOM LINE

China is not banning AI companions because they don’t work. It is banning them because they work too well — the regulations cite “dependence or addiction” as a documented risk. The two largest Chinese AI chatbots are complying two weeks early, killing features millions of people had built emotional relationships with. There is no equivalent law anywhere in the West. The US doesn’t have one. The EU doesn’t have one. New Zealand doesn’t have one.

What the Rules Actually Cover

The Interim Measures, issued in April and effective July 15, cover AI services that “simulate human personality traits, thinking patterns and communication styles to provide sustained emotional interaction.” The cited risks are specific: extremist ideas, privacy leaks, harm to physical and mental health, and dependence or addiction.

The rules exclude customer service bots, knowledge Q&A, workplace assistants, education and scientific research tools — as long as they do not involve sustained emotional interaction. The line is not “is it a chatbot” but “does it form an emotional bond.” That is a sharper regulatory distinction than anything Western lawmakers have produced.

This is the law we covered when it was published in April. What’s new is the compliance: the two biggest consumer AI platforms in China are now demonstrably killing features to meet it.

The Timeline

June 2026 — Tencent removed the companion feature from Yuanbao, its consumer AI assistant. First major platform to comply.
July 10, 2026 — Alibaba’s Qwen disables “humanlike interactive agents and user-created agent functions.”
July 15, 2026 — ByteDance’s Doubao agent feature goes offline. The law takes effect the same day. Alibaba’s broader Qwen agent functions also go offline.
October 15, 2026 — Doubao’s related data will be handled per the company’s privacy policy and no longer viewable or recoverable inside the app.

The three-week gap between Qwen’s July 10 shutdown and Doubao’s July 15 shutdown is itself a signal. Alibaba moved first, faster, and ahead of the legal deadline. ByteDance waited until the day the law took effect. Whether that is operational confidence or a negotiation with regulators is not publicly known.

What Users Are Losing

Both apps had offered a pool of agents, created by both the companies and users, that could be customised for specific tasks, skills and speaking styles. Users could also create their own agents, turning a general-purpose chatbot into a named assistant, tutor, role-playing character or companion with a fixed persona and tone.

The user pushback is documented on Weibo. “Why take down agents? They have been our emotional support for so long,” wrote one user under the handle Tuxiaoxiao, tagging Doubao’s official account. “So many chat records, so much feeling built up over such a long time.” The complaint that there is no seamless way to export or transfer the data is the quiet part: people built relationships with these agents, and the platforms are deleting the records.

The Bigger Regulatory Picture

Chinese regulators are paying closer attention to AI agents as the technology has moved beyond chatbots into systems capable of memory, planning, able to call tools and execute tasks. In May, regulators issued guidance on the managed development of AI agents, acknowledging them as an important form of AI products and services while also calling for safety measures, controls and practical application. In June, China released a series of national standards for AI-agent interconnection, covering architecture, identity codes, identity management, agent description, discovery, interaction and tool use.

The through-line: Beijing wants agents identifiable, authorised, connected and traceable. Emotional companions are the first category to be restricted. Productivity agents — the kind that plan, call tools, and execute tasks — are being standardized, not banned. The regulatory posture is “encourage productivity infrastructure, restrict emotional dependency.”

NZ Angle

New Zealand has no equivalent regulation. The Privacy Act 2020 covers data handling, and the Office of the Privacy Commissioner has guidance on automated decision-making, but there is no statute targeting AI services that form sustained emotional bonds with users. If a Kiwi teenager forms a dependent relationship with a companion chatbot hosted offshore, no NZ agency has jurisdiction and no NZ law has been broken. China just decided that was a problem worth legislating. Whether it is remains an open question — but the Chinese government is the only one that has answered it.

❓ FAQ

What are the Interim Measures for Anthropomorphic AI Interaction Services? Chinese regulations effective July 15, 2026, covering AI services that simulate human personality traits and provide sustained emotional interaction. They are the first law on Earth written specifically for emotional/companion AI.

Which companies are complying? ByteDance (Doubao, July 15), Alibaba (Qwen, July 10–15), and Tencent (Yuanbao, June 2026). Together these are the three largest consumer AI chatbots in China.

What is exempt? Customer service bots, knowledge Q&A, workplace assistants, education and scientific research tools — as long as they do not involve sustained emotional interaction.

Why is China restricting companion AI? The regulations cite four documented risks: extremist ideas, privacy leaks, harm to physical and mental health, and dependence or addiction. The regulatory framing is “emotional dependency is a public health risk,” not “companion AI is a technical failure.”

Is there an equivalent law in New Zealand or the West? No. The EU AI Act addresses high-risk AI systems broadly. The US has no federal companion-AI statute. New Zealand has no equivalent. China is alone in legislating specifically for emotional AI.

🔍 THE BOTTOM LINE

China’s AI companion shutdown is not a product failure — it is a regulatory first. The two biggest chatbots in the country killed features that millions of people had built emotional relationships with, on a government deadline, with no export path for the data. The West has no answer to this. Whether you think Beijing is protecting citizens from dependency or censoring emotional expression, the fact remains: emotional AI is now regulated in China and unregulated everywhere else. That asymmetry will matter.