China muzzles its humanoid AI agents, a disguised admission of weakness

Introduction: when Beijing unplugs what it once encouraged

A countdown that started on July 5

Since July 5, 2026, two of China's biggest tech giants, ByteDance and Alibaba, have begun disabling the custom agent features of their respective artificial intelligence assistants, Doubao and Qwen. This decision comes just days before a new Chinese regulation strictly governing human-like AI interaction services takes effect on July 15, 2026.

According to reporting from the South China Morning Post, confirmed by Bloomberg, Doubao, China's most popular AI app with about 345 million monthly users, warned its users that its custom agent feature would stop working by that deadline, directing users to a separate app called Maoxiang to continue that kind of use.

A last-minute compliance maneuver

This regulation, officially titled the "Provisional Measures for the Administration of Anthropomorphic Artificial Intelligence Interaction Services," was jointly published in April 2026 by the Cyberspace Administration of China and four other government agencies, including the Ministry of Industry and Information Technology and the Ministry of Public Security.

The rushed withdrawal of these features by the two tech giants, just days before the regulatory deadline, illustrates a reality the West would be wrong to ignore: even China's most powerful tech companies remain, in the final analysis, at the mercy of a central political power able to rewrite the digital rules of the game overnight.

What the new Chinese regulation precisely bans

Anti-addiction mechanisms now mandatory

The new regulation requires providers of AI services simulating a human personality to build in anti-addiction mechanisms, mandatory usage notifications after two hours of continuous interaction, and instant exit options letting users immediately end any interaction with the artificial agent.

The text explicitly bans any attempt at emotional manipulation aimed at pushing a user toward an unreasonable decision, as well as any form of excessive emotional dependency that could harm a user's real interpersonal relationships, a concern increasingly documented worldwide as companion chatbots proliferate.

Special protection for minors and the elderly

The regulation flatly bans offering "virtual partner" or "virtual parent" services to minors, a provision reflecting clear government concern about the psychological impact of these emotionally immersive technologies on the country's most vulnerable young users.

People under 14 will need explicit consent from a legal guardian before accessing this kind of service, while elderly people will receive enhanced warnings and dedicated support, two population groups Chinese authorities clearly consider particularly exposed to the risks of these technologies.

The fast, coordinated response from Chinese tech giants

Doubao and Qwen, an almost perfect synchronization

According to BigGo Finance, the announcements from ByteDance and Alibaba came nearly simultaneously, a timing several tech-sector analysts interpret as a coordinated regulatory compliance maneuver rather than a purely commercial coincidence between two direct competitors in China's consumer artificial intelligence market.

Qwen announced it would disable its humanoid conversational agents and user-created agent features starting July 10, 2026, ahead of a complete shutdown of all agent services on July 15, a two-stage rollout that leaves little room for last-minute improvisation.

Tencent had already paved the way in June

Tellingly, Tencent, through its Yuanbao app, had already withdrawn a similar feature back in June 2026, getting ahead of the regulation's effective date rather than waiting until the last minute as its rivals ByteDance and Alibaba did.

This early move by Tencent suggests the entire Chinese tech industry had, since the regulation was published in April, precise knowledge of what would become illegal or heavily restricted starting mid-July, making the delay taken by ByteDance and Alibaba before acting all the more notable.

What this episode reveals about Chinese AI governance

China as a pioneer in regulating affective AI

Several international experts, including the law firm Linklaters and the publication China Law Translate, note that China is positioning itself as the first jurisdiction in the world to adopt legislation this detailed and binding specifically for anthropomorphic interaction services, getting ahead of the European Union and the United States on this specific front.

This Chinese regulatory head start on a topic as sensitive as affective AI deserves to be recognized objectively, even though it fits within a broader logic of social control that fundamentally distinguishes the Chinese approach from any equivalent Western attempt at technology regulation.

A control that extends well beyond user protection alone

The Chinese regulation also bans using sensitive conversation data to train future AI models without explicit consent, a provision that, on paper, protects users' privacy, but comes paired with requirements for compliance with core socialist values imposed on all content generated by these systems.

This dual nature of the regulation, both individually protective and ideologically restrictive, nicely illustrates the specificity of the Chinese model of technology governance, where citizen protection and political control of digital content always advance hand in hand rather than as separate goals.

The bigger stakes in the global artificial intelligence race

China regulating while innovation keeps moving

It would be wrong to think this wave of regulation is slowing China's overall technological ambition: on the contrary, this same period saw the publication of joint guidelines by the Cyberspace Administration, the National Development and Reform Commission, and the Ministry of Industry to accelerate the development and deployment of artificial intelligence agents under the national "AI Plus" action plan.

This dual approach, restrictive on uses deemed socially risky but firmly permissive on general technological innovation, illustrates China's strategy of channeling rather than slowing the development of artificial intelligence, an important nuance Western commentators sometimes tend to overlook.

The West must maintain a strategic edge

For the West, this episode is a reminder of a fundamental strategic urgency: the race for artificial intelligence isn't limited to computing power or model quality alone, it also includes the ability to establish coherent regulatory frameworks that protect citizens without stifling innovation, a balance neither Washington nor Brussels has yet perfectly found.

Facing a China that methodically advances on both fronts at once, regulation and innovation, Western democracies cannot afford to remain paralyzed by endless debates while Beijing patiently builds its technological and normative advantage on a global scale.

The global precedents shedding light on China's decision

Italy and the United States had already opened the debate

China isn't entirely inventing this regulatory territory: Italy had already banned the virtual companion app Replika in 2025 for non-compliance with European data protection rules, while several American states, including California, have passed laws requiring clear warnings when users interact with a chatbot rather than a real human.

What sets the Chinese approach apart, however, is the scale of the market affected: with hundreds of millions of active users on apps like Doubao, no Western jurisdiction has yet had to impose a regulation of this scope on such a massive user base in so little time.

The economic fallout for Chinese AI companies

A significant compliance cost for the tech giants

According to analysts cited by Reuters, complying with these new rules will force ByteDance and Alibaba to redirect considerable engineering resources toward overhauling their user interfaces, building robust age-verification systems, and developing real-time notification mechanisms, all costs that will directly weigh on their margins in an already highly competitive sector.

This sudden regulatory burden comes just as these companies are investing massively in the race for large language models against rivals like DeepSeek and Moonshot AI, a dual financial and technological pressure that could temporarily slow their pace of innovation on other fronts.

What this means for Chinese users day to day

A user experience abruptly transformed

For the millions of daily users of Doubao and Qwen, this transition means having to relearn how to use apps around which some had built deeply ingrained habits, with some reporting on Chinese social networks like Weibo a sense of loss at the sudden disappearance of custom agent features they had grown attached to.

Chinese authorities are counting on the replacement app Maoxiang to absorb this demand, but nothing guarantees the migration will be friction-free, with several users fearing the loss of personal data or conversation histories built up over months of intensive use.

Conclusion: a lesson in power more than a lesson in caution

What Beijing is really demonstrating through this episode

Ultimately, this rushed withdrawal of custom agent features by ByteDance and Alibaba demonstrates less a newly discovered ethical caution on the part of these companies than a show of force by China's central power over its tech industry, capable of imposing major changes within weeks on products used by hundreds of millions of people.

This capacity for fast, coordinated execution, however impressive in terms of administrative efficiency, must be understood for what it really is: an exercise of political authority rather than a governance model to be blindly imitated by Western democracies, which must find their own balanced answers to these same technological challenges.

The call for a more coherent Western response

As China imposes rules on its artificial intelligence industry with formidable speed, the West must accelerate its own smart regulatory efforts, without sacrificing the fundamental freedoms that distinguish our democratic societies from the centralized control model practiced in Beijing.

It is this delicate balance between protecting citizens, preserving individual freedoms, and maintaining cutting-edge technological innovation capacity that will ultimately determine which governance model for artificial intelligence proves most durable and most worth following in the long run by the rest of the world.

By Maxime Marquette, columnist

Columnist's transparency note

Who I am and my acknowledged biases

I sign this essay as an openly pro-Western columnist, convinced that liberal democracies must defend a model of technology governance fundamentally different from the one practiced by China, while objectively acknowledging certain elements of regulatory rigor present in the Chinese regulation analyzed here.

My analysis relies on established journalistic and legal sources, notably the South China Morning Post, Bloomberg, Reuters, and law firms specializing in Chinese technology law such as Linklaters and China Law Translate.

What I don't know, and my method

I cannot predict with certainty how this regulation will be applied in practice over the long term, nor whether other countries will draw on it to regulate their own conversational artificial intelligence services. My method is to rigorously separate verifiable regulatory facts from geopolitical and editorial interpretations, always identified as such in this text.

Sources

Primary sources

Business Standard — ByteDance, Alibaba pull AI companion features as China tightens rules, July 6, 2026

BigGo Finance — Doubao and Qianwen simultaneously axe consumer AI agents ahead of China's new anthropomorphic AI rules, July 4, 2026

Secondary sources

China Law Translate — Provisional Measures on the Administration of Human-like Interactive Artificial Intelligence Services, April 10, 2026

Bird & Bird — China's New Regulations on AI Anthropomorphic Interactive Services, June 17, 2026

China AI Dispatch — The Off Switch, July 4, 2026

Global Times — Chinese LLMs Doubao, Qwen to shut down personalized AI agents on July 15, July 5, 2026