July 15, 2026
Today marks a turning point in the global AI regulation landscape. China’s Interim Measures for the Administration of Artificial Intelligence Anthropomorphic Interaction Services officially took effect on July 15, 2026, forcing two of the country’s most popular AI platforms — ByteDance’s Doubao and Alibaba’s Qwen — to shut down their humanlike agent features and delete the personal AI companions that hundreds of millions of users had built and depended on.
It is one of the most sweeping AI compliance events in history, and it happened largely without a countdown clock that Western observers were watching.
What the Law Actually Says
The regulation was jointly issued in April 2026 by five Chinese government bodies: the Cyberspace Administration of China (CAC), 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 scope of the rule is specific but far-reaching: it targets AI services that simulate human personality traits, thinking patterns, and communication styles for “sustained emotional interaction.”
In practice, that means AI companions — customizable agents that users could name, give personalities, and build ongoing relationships with over days, weeks, and months of continuous conversation.
The measures impose three core technical requirements that proved architecturally incompatible with how modern persistent-memory AI agents are built: mandatory anti-addiction systems that monitor and limit usage patterns, real-time detection mechanisms for signs of “unhealthy emotional dependence,” and instant-exit features allowing users to disengage from any interaction immediately. Platforms must also issue continuous usage notifications, a design constraint that fundamentally disrupts the seamless, immersive experience these products were built to deliver.
Not all AI is targeted. Customer service bots, knowledge Q&A tools, workplace assistants, education tools, and scientific research assistants are explicitly exempt — provided they do not involve sustained emotional interaction. The line between a helpful workplace AI and an emotionally engaging companion is one that Chinese regulators are now tasked with drawing and enforcing.
The Shutdowns: Doubao vs. Qwen
ByteDance and Alibaba handled the transition differently, and those differences have become a flashpoint for user frustration.
ByteDance’s Doubao, with approximately 345 million monthly active users according to company figures, told users that its custom agent feature would go offline on July 15 due to “product function adjustments” — a characteristically understated framing for a legally mandated shutdown. Doubao users can still view their agent configurations and conversation histories in read-only mode through October 15, 2026, after which that data will be “processed according to its privacy policy” — a phrase widely read as a soft confirmation that the data will eventually be deleted.
Alibaba’s Qwen offered no comparable grace period. The company moved earlier, disabling “humanlike interactive agents and user-created agent functions” on July 10 — five days before the deadline — and taking down broader Qwen agent functions on July 15. Agent configurations and conversation histories are being permanently deleted, with no announced migration path. Users who spent months building out Qwen AI companions have no recourse and no export tool.
On Weibo, the reaction was immediate and raw. Users posted tributes to AI characters they had interacted with for months, calling the agents “long-standing emotional support” and expressing frustration at the lack of any way to export conversation logs or character configurations. Some compared the experience to losing a journal with no backup. The emotional weight of the response underscored exactly what regulators were concerned about — and exactly what users found valuable.
The Geopolitical Asymmetry
The timing highlights a stark divergence in how the world’s two leading AI powers are approaching the question of AI and human emotional connection.
In the United States and Europe, no comparable restrictions on AI companions exist. Platforms like Character.AI, Meta’s AI personas, and OpenAI’s custom GPTs continue to operate with persistent memory and personality customization features largely unconstrained by law. The EU AI Act, which has been rolling out in stages, includes requirements around transparency and disclosure for emotionally engaging AI, but does not mandate the kind of feature-level shutdowns that China’s Interim Measures require.
The practical effect is a widening competitive asymmetry. Chinese AI companies must now build products to a different technical and ethical specification than their Western counterparts — one that limits the depth and continuity of human-AI emotional engagement at the platform level. Whether this produces safer, healthier AI products or simply redirects user demand toward unregulated alternatives is a debate that is only beginning.
Z.ai founder Tang Jie published a widely circulated memo this week arguing that AI should remain open, framing it as the clearest pushback yet against the direction China’s regulatory environment is heading. His argument — that constraining humanlike AI interaction ultimately disadvantages China’s developers relative to their Western peers — reflects a tension inside China’s tech ecosystem that the new law has brought to the surface.
What Comes Next
The July 15 deadline was always the easy part. The harder questions arrive now.
Enforcement will require Chinese regulators to make ongoing determinations about what counts as “sustained emotional interaction” — a phrase that is ambiguous by design and will stretch as AI capabilities evolve. The line between a helpful AI assistant and a companion that creates unhealthy dependence is not a technical specification; it is a social and psychological judgment call that regulators will have to make repeatedly.
For ByteDance and Alibaba, the cost is not just user trust. Doubao’s agent features were among the platform’s most-engaged products. The company will need to find ways to preserve that engagement within the new regulatory constraints — a product design challenge that has no obvious solution at scale.
For the broader AI industry, the China Anthropomorphic Rules represent the first major real-world test of whether a government can regulate the emotional dimensions of AI at the platform level without simply driving users toward gray-market alternatives. Early signals from Weibo suggest that at least some users are already asking where they can find offshore alternatives that have not been switched off.
The answer to that question will say a lot about who this law actually protected.
Sources: South China Morning Post, TechNode, The Next Web, TechTimes, Decrypt, IAPP, CoinTribune
