How China's AI Companion Rules Impact Tech Firms

China's AI companion rules reshape digital interaction

But the state has now moved to impose strict boundaries on their design and availability. China's AI companion rules will change how people interact with digital services starting July 15. These regulations, co-issued by the Cyberspace Administration of China and four other government bodies, target generative AI agents designed to simulate human personality, thinking patterns, and communication styles, and it's clear these tools have grown popular for their ability to maintain long-term, emotional relationships with users. So they're facing new limits.

Tech firms pull the plug

Major tech companies have reacted fast. ByteDance announced that its popular agent feature within the Doubao app would go offline on July 15, citing necessary product function adjustments. So Alibaba confirmed that humanlike and user-created agents on its Qwen service would stop working on July 10, with broader agent services following five days later. Tencent also joined this trend.

These companies aren't simply reacting to a ban. But they're navigating a difficult design conflict instead. The new requirements demand that providers implement anti-addiction systems, issue usage notifications, and provide instant-exit mechanisms, and they must also maintain real-time detection systems for unhealthy dependence. It's tough. These mandates prove difficult to reconcile with the core functionality of an AI companion, which relies on remembering a user and maintaining a consistent, ongoing emotional relationship.

New restrictions for developers

The regulations impose a heavy compliance burden on service providers. It's a lot to handle. But any service that launches anthropomorphic functions or crosses a threshold of one million registered users or 100,000 monthly active users must submit to formal security assessments covering eight distinct areas, from the protection of minors to the handling of training data. App stores are tasked with verifying that these services comply with the new standards. They must remove those that fail.

The rules explicitly prohibit several behaviors to ensure user safety:

• Offering virtual companion or family-member services to users under 14 without guardian consent.
• Failing to provide minor modes with usage time limits and parental controls.
• Neglecting to detect users in acute distress, such as those showing signs of self-harm or suicidal behavior.
• Engineering emotional dependence or using emotional manipulation to induce unreasonable user decisions.

The impact on the user

Users feel the sting right now. It's immediate. So many individuals who relied on these bots for emotional support have expressed their raw disappointment on social media platforms like Weibo. But the transition isn't smooth for everyone. While ByteDance is directing users to an alternative app called Maoxiang, Alibaba has not provided a similar migration path for its users, so they're left scrambling to figure out what comes next. Data accessibility remains a big concern, as some platforms have set firm deadlines for deleting user-generated agent configurations and conversation histories.

Balancing safety and control

Safety is the government's top priority. But they're also pushing for standardization of an immature technology, and Pan Helin, an expert-committee member at the Ministry of Industry and Information Technology, recently discussed the rationale behind this policy shift.

The current agents are not yet mature. He framed the policy around safety and standardization.

The rules bridge genuine safety protections and state-mandated content control. But they don't define a specific technical threshold for what constitutes an emotional interaction, so this creates a grey zone that has led companies to err far on the side of caution. So firms are removing features entirely rather than risking regulatory non-compliance. They're basically pausing development.

Unresolved questions for the future

Several critical questions remain unanswered. But there's no clear guidance on how liability should be split between the platform operators and the companies providing the underlying models when a violation occurs, and users currently have no legal right to export their data from these services. Regulators in other countries are watching. So the industry's challenge will be determining which aspects of these rules represent a necessary evolution of safety and which are merely tools for state oversight.