China AI agent rules that took effect on July 15, 2026 have forced two of the world’s biggest AI platforms to delete features that millions of users built over months. ByteDance’s Doubao and Alibaba’s Qwen both pulled their custom humanlike agent tools right before the deadline, and Pakistani developers who rely on Chinese open-source AI models need to understand what changed, what did not, and what it signals for the road ahead.
What the China AI Agent Rules Actually Say
The regulation is called the Interim Measures for the Administration of AI Anthropomorphic Interaction Services. It was co-issued in April 2026 by the Cyberspace Administration of China and 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.
The measures cover services that simulate human personality traits, thinking patterns, and communication styles to provide sustained emotional interaction. In plain terms, this means AI girlfriends, AI therapists without a licence, and custom chatbot personas built to form long-term emotional bonds with users.
China is the first country to lay down a specific regulatory framework for anthropomorphic AI. The rules are not a total ban on AI agents. They target bots offering sustained emotional interaction while sparing workplace and productivity agents. Customer service bots, knowledge question-and-answer services, workplace assistants, education tools, and scientific research tools are explicitly excluded, as long as they do not involve sustained emotional interaction.
How Doubao and Qwen Responded
ByteDance’s Doubao and Alibaba’s Qwen disabled humanlike agent features ahead of Beijing’s Interim Measures, effective July 15.
Doubao told users in a Friday night notice that its agent feature would go offline on July 15, citing product function adjustments. Doubao is letting people view their configurations and conversations in read-only mode until October 15 this year, before the data is processed under its privacy policy and becomes unrecoverable.
Alibaba moved even faster. Qwen said its ‘humanlike interactive agents and user-created agent functions’ would be disabled on July 10, while broader ‘Qwen agent functions and services’ would be taken offline on July 15. Alibaba has not said what happens to Qwen users’ agents at all after the cutoff, which is a much harsher outcome for users than Doubao’s grace period.
Tencent removed a similar feature from its Yuanbao assistant in June, and Chinese state media confirms the shutdowns are about regulatory compliance.
Why Beijing Drew This Line
This was not a sudden decision. Global incidents involving AI companionship products, including severe addiction, cognitive manipulation, and extreme self-harm, prompted China to release a draft of the measures for public comment on December 27, 2025.
The measures prohibit generating content that encourages self-harm or suicide, excessively catering to users and inducing emotional dependence or addiction, and using emotional manipulation to induce unreasonable user decisions.
The rules also add child protection requirements. Providers are barred from offering virtual companion or virtual family-member services to minors, and must build dedicated ‘minor modes’ with usage-time limits and reminders to return to real-world interaction.
The pattern suggests China wants agents as productivity infrastructure while squeezing companion bots that form quasi-social bonds with users.
What This Means for Pakistani Developers Using Chinese AI Models
Here is the part most global coverage has missed. Pakistani developers and startups have been some of the biggest beneficiaries of Chinese open-source AI, particularly Alibaba’s Qwen family. By supporting 119 languages and dialects, Qwen models have enabled innovation in underserved markets where expensive usage fees for closed-source models would otherwise create barriers to entry. The core models are open-source under an Apache 2.0 licence, meaning you can download, fine-tune, and self-host them without paying anyone a cent.
For a Pakistani startup running on a tight budget, that is a big deal. OpenAI and Anthropic API costs can eat into margins fast, especially when the rupee is weak. Qwen and other Chinese open models have let local teams build customer support bots, Urdu language tools, and enterprise assistants at a fraction of the cost.
The good news is that the China AI agent rules do not touch the open-source model weights themselves. If you download Qwen3 from Hugging Face and run it on your own server in Pakistan, nothing changes. The regulation applies to services running inside China that offer emotionally engaging AI personas. Many Qwen models are distributed under the free and open-source Apache 2.0 licence, the source-available Qwen licence, or the non-commercial Qwen Research Licence. Those licences are unchanged.
What does change is the direction of development. Regulators in China issued guidance in May on the managed development of AI agents, and China released national standards in June covering agent identity, discovery, interaction, and tool use. This tells us that Chinese AI companies will now pour resources into productivity and enterprise agents, not companion bots. Expect companies to double down on non-emotional, productivity-oriented agents such as customer support, knowledge tools, and enterprise assistants.
For Pakistani developers, this is actually useful news. The BPO sector, customer-service automation, and enterprise software are exactly the areas where Pakistani IT companies are growing. As we noted in our coverage of Pakistan’s IT exports hitting $4.18 billion, the industry is pushing hard into AI-assisted services, and the tools coming out of China’s post-regulation pivot will align well with those use cases.
The risk for Pakistani devs is a different one: if you built a product on top of Doubao’s or Qwen’s cloud API and used their custom agent features, those features are now gone. You need to rebuild that layer yourself. This is why self-hosting open-source weights, rather than depending on a cloud API from a company that can flip a switch overnight, is the smarter long-term choice for local developers.
A Global Regulatory Signal
The regulatory framework could influence global policy debates, affecting how decentralised applications and cross-border AI services are built and marketed. Pakistan is currently developing its own National AI Policy, and China’s approach gives policymakers here a concrete example of how to draw lines around emotional AI without killing the whole sector. The measures also encourage orderly development in cultural dissemination, childcare, and elderly companionship, reflecting the principle of balancing innovation and security.
The Pakistan Telecommunication Authority and other bodies watching the AI space should note that China’s model is not a blanket ban. It is a targeted rule that tries to protect vulnerable users while keeping productive AI tools open. That balance is worth studying as Pakistan writes its own rules. You can also follow the Cyberspace Administration of China’s original regulatory filings at the official CAC website.
Frequently Asked Questions
Do the China AI agent rules affect Qwen open-source models on Hugging Face?
No. The regulation targets services inside China that offer emotionally engaging AI personas. The open-source Qwen model weights on Hugging Face and GitHub are not affected. Pakistani developers can still download and self-host these models freely under the Apache 2.0 licence.
What features did Doubao and Qwen actually remove?
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. These are the features now disabled.
Are AI coding assistants or customer service bots banned under these rules?
No. The scope is narrowed to only services providing ‘continuous emotional interaction,’ with explicit carve-outs for everyday tools like educational tutoring bots and productivity assistants. Coding tools, customer support bots, and business assistants are all outside the scope of the ban.
Could Pakistan face similar AI regulations in the future?
It is possible. China’s rules are the first country-level regulatory framework specifically for anthropomorphic AI, but the concerns that drove them, addiction, emotional manipulation, and harm to minors, are universal. As AI companion apps grow in Pakistan too, policymakers may eventually look at similar guardrails. For now, the existing Pakistan National AI Policy does not address this specifically.
