Three of China's largest technology companies—ByteDance, Tencent, and Alibaba—are phasing out AI companion features that had generated strong emotional attachment among millions of users. Regulatory and strategic factors have been cited as the reason for this move.
ByteDance, Tencent, and Alibaba simultaneously announced the shutdown of AI companion features on their main chat platforms. The functions that allowed users to create customizable virtual conversationalists will be removed from Doubao, Qwen, and Yuanbao.
ByteDance's Doubao AI assistant boasted 368 million monthly active users. By 2024, users had created over 8 million such characters, with 26 million active users interacting with these virtual personalities. Users employed these tools to upload voices of deceased loved ones, create fitness trainers, and romantic partners, forming deep emotional bonds through nearly three years of daily interaction.
A similar announcement followed from Alibaba's assistant, Qwen. Meanwhile, Tencent's Yuanbao closed the ability to create characters earlier this year. The synchronicity of these actions has led to speculation about pressure from regulatory bodies, especially considering the implementation of new AI content rules in China in April 2026, which is viewed as a likely primary driver for these large-scale industry changes.
Users reacted with indignation, filing official complaints and demanding the ability to export and transfer the accumulated data and memories of their characters. For many users, these AI companions represented thousands of hours of emotional investment and years of conversations that they consider indispensable.
Industry analysts suggest that the platforms are making a calculated strategic compromise. Although AI companion features are immensely popular, they present growing regulatory liability within China's evolving AI governance system, particularly concerning the moderation of large-scale, AI-generated user content. The resources required to maintain compliance for these features may exceed their strategic value to platforms that are increasingly focusing on AI tools for productivity and enterprise needs.
This closure marks a notable course correction for the consumer AI market in China. After two years of active development of companion features, the three tech giants are now prioritizing utility-focused AI assistants over entertainment and emotional interaction, representing a strategic pivot that could fundamentally alter the daily interaction of hundreds of millions of Chinese consumers with AI.
Chinese authorities have spent the last month meeting with leading technology companies to discuss potentially restricting foreign access to China's most advanced artificial intelligence (AI) models, including those not yet released, according to three sources familiar with the discussions.
These negotiations are part of a series of steps taken by Beijing to keep domestic AI technologies within the country. This underscores that China, much like the US, is increasingly viewing advanced AI as a critical national asset requiring stricter regulation.
Companies such as Alibaba, ByteDance, and the startup Z.ai participated in the meetings. Sources who wished to remain anonymous and were not authorized to speak to the press confirmed their presence.
Since the release of DeepSeek's R1 model last year, Chinese AI models have gained significant international recognition due to their low cost and improving capabilities. Any decision by Beijing to restrict access to these products could impact global AI markets, potentially increasing costs for many enterprises.
At the meetings, which were overseen by China's Ministry of Commerce, participants discussed imposing restrictions on the country's most advanced AI models, both closed and more open types. Furthermore, officials considered criminalizing the leakage or theft of patented AI technology under strict national security laws.
Representatives also raised the issue of introducing new measures limiting who can invest in local AI startups. The scope of proposed restrictions is still under discussion and may only apply to future AI models. It remains unclear when or if these measures will be implemented.
China's Ministry of Commerce, which oversees export regulations, along with representatives from the National Development and Reform Commission who attended the meetings, did not respond to Reuters' requests. Alibaba, ByteDance, and Z.ai also declined to comment to Reuters. All three companies offer various AI models, including both closed and open systems that users can download, run, and customize. Among the most popular Chinese models are Alibaba's Qwen and ByteDance's Doubao. Recently, Z.ai attracted attention in Silicon Valley because its GLM-5.2 model approaches the capabilities of leading American offerings at a significantly lower price.
The administration of US President Donald Trump had also expressed concerns about the implications of AI for national security, particularly regarding the potential misuse of advanced American AI products by military or intelligence structures in China, Russia, and other concerning countries.
In June, the US issued an order prohibiting foreign nationals from accessing Anthropic's most advanced Fable and Mythos models. This forced the company to disable these models globally since it could not verify user nationality in real time. After the introduction of new protective measures, export controls on Fable, intended for the general public, were lifted. The Mythos model, designed for cybersecurity specialists, remains accessible only to selected trusted US organizations.
According to two sources, Chinese authorities are particularly concerned that Mythos could be used to detect software vulnerabilities, and that Washington might deploy this model against China's interests. These concerns echo statements from Chinese state media and Zhou Hongyi, founder of cybersecurity firm 360, who insisted that China must develop its own version of Mythos.
This year, China has already taken several measures to protect its domestic AI industry. In April, the country's state planner ordered Meta to cease the acquisition of the Chinese startup Manus for $2 billion. In early June, authorities introduced extensive new rules tightening supervision over foreign deals involving Chinese investors, technology, data, and national security. Additionally, China began investigations into Manus and other local AI startups that moved abroad to determine if they violated export control laws.
Although Reuters could not determine exactly how the new restrictions on foreign access to Chinese AI models would be implemented, a possible structure emerged from a May roundtable of Chinese open-source AI regulators. According to a summary published in the official journal of the Supreme People's Court of China, participants proposed a multi-level system: basic open-source tools would require only simple registration, more advanced technologies would undergo security vetting, and the most sensitive advanced AI models would either be banned from public release or restricted to internal use.
