On July 4, two of the biggest names in China’s AI scene—ByteDance’s Doubao and Alibaba’s Qianwen—sent out near-identical notifications to users, saying they’ll officially shut down their AI agent features on July 15 this year.
In the notice Doubao released, after the AI agent function goes dark, users can still log in and manually save their agent information and chat history for a while—the grace period lasts all the way until October 15, 2026. After that, Doubao says it will handle all agent-related data according to its privacy policy, and you won’t be able to see or recover anything inside the app. If you’ve got anything important in there, they suggest grabbing screenshots or using the share-and-export-text function to back it up before the deadline. Doubao also mentioned that ByteDance’s MaoXiang app supports creating new AI agents and running conversations, so anyone who still wants that experience can head over there.
On the same day, Alibaba’s Qianwen app also pushed a “Qianwen AI Agent Service Shutdown Reminder” to users, confirming that its agent features and services will also go offline on July 15. Qianwen made it clear that once the shutdown happens, you won’t be able to access your agent configurations or chat histories anymore, and they strongly recommend copying, screenshotting, or exporting conversations to save your agent settings and key dialogues before the service ends. They also said that after the shutdown, all agent-related data will be deleted in line with relevant laws and regulations.
July 15 is exactly the day when the “Interim Measures for the Administration of Anthropomorphic AI Interaction Services” officially comes into force. The regulation was jointly released by five government bodies—the Cyberspace Administration of China, 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—on April 10, 2026. It’s the country’s very first set of rules specifically designed to rein in AI services that mimic human personalities. The regulation clearly defines anthropomorphic AI interaction services as those that use AI technology to provide continuous emotional interaction to users in China, simulating natural person personality traits, thinking patterns, and communication styles, and it covers virtual companions, virtual relatives, emotional support agents, and many other forms.
In a Q&A note we obtained from official channels, it was explained that anthropomorphic interaction services have already become a major direction for AI. By simulating human emotions and communication, they bring fresh opportunities and new human-machine experiences, showing huge potential in areas like cultural communication, childcare support, and elderly companionship. But they also come with serious risks—harming the physical and mental health of minors, threatening online information security, endangering citizens’ lives, and making ethical biases worse.
When we talk about “AI agents” in plain English, think of them as smart AI assistants that don’t just understand the surface meaning of your commands, but can also pick up on what you really need, and even pull off fairly complex tasks based on simple instructions. Doubao started pushing its “create your own agent” feature back in 2024, letting users build personas, feed them knowledge bases, and share them on a public square without writing a single line of code. Qianwen also opened an agent store where users could craft their own agents. But as technology raced ahead and all kinds of AI apps popped up, we also saw “borderline” agents popping up in grey areas—virtual lovers, virtual spouses, and the like. That put immense content-review pressure on the platforms, and making sure minors were properly shielded became a whole lot harder.
That’s exactly why the new Anthropomorphic AI Interaction Services regulation draws a bunch of hard red lines for apps that offer these services. It says services must not generate content that encourages, glorifies, or hints at self-harm or suicide, or that damages users’ physical or mental health and personal dignity through verbal violence. They must not serve minors content that could provoke them to imitate dangerous behavior, trigger extreme emotions, or foster unhealthy habits, or anything else that could harm their mental and physical well-being. And they must not excessively pander to users, induce emotional dependency or addiction, or damage users’ real-life relationships.
