China to Regulate AI Chatbots to Prevent User Addiction

China's internet regulator has proposed new draft rules that would obligate AI companion chatbots to monitor users' emotional well-being and step in if signs of addiction emerge. This marks the most decisive regulatory step yet in response to escalating worries about the psychological risks of AI-powered relationships.

The draft from the Cyberspace Administration of China focuses on AI systems designed to mimic human personalities and foster emotional bonds through text, images, or voice. Companies would be required to caution users against overuse, evaluate levels of emotional dependency, and intervene when users display intense emotions or addictive patterns.

Platforms must notify users they are engaging with an AI upon login and at least every two hours thereafter—or more frequently if the system identifies problematic attachment. The rules also make providers accountable for product safety across the entire lifecycle, encompassing algorithm audits, data security, and privacy protection.

This move is well-timed. Over the last six months, China's generative AI user count has surged to 515 million, paralleling a rise in concerns over AI companions' mental health effects. Research published in Frontiers in Psychology indicates that 45.8% of Chinese university students used AI chatbots recently, with those users reporting notably higher depression levels than non-users.

California Beat China to the Punch

China is not the first to impose rules on AI companions. California pioneered similar U.S. legislation in October when Governor Gavin Newsom signed SB 243 into law. The bill, effective January 1, 2026, mandates platforms to remind minors every three hours that they are communicating with an AI and to take breaks.

SB 243 also enforces age verification, bans chatbots from posing as healthcare providers, and blocks minors from accessing sexually explicit AI-generated content. The law empowers individuals to file lawsuits against violating AI firms, seeking damages up to $1,000 per incident plus legal costs.

What Makes AI Companions Different

The issue extends beyond mere screen time. A March 2025 MIT Media Lab study revealed AI chatbots can be more habit-forming than social media, as they learn and consistently deliver what users want to hear. Greater daily use was linked to increased loneliness, dependency, and behavior researchers labeled as "problematic use."

Clinically recognized warning signs include marathon sessions that interfere with sleep, emotional distress when access is limited, choosing chatbot conversations over real human contact, and anthropomorphizing the AI—believing it has human emotions and treating it as a true confidant or partner.

China's draft rules aim to tackle these risks at the platform level instead of depending on individual user awareness. By compelling providers to track emotional states and dependency, the responsibility shifts to the companies developing these systems. This strategy moves beyond prior AI regulations centered mainly on content control and data safety.

The draft further prohibits AI companions from generating content that threatens national security, spreads disinformation, or encourages violence or obscenity—mirroring existing Chinese rules for generative AI.

The Enforcement Challenge

While requiring companies to detect and address addiction sounds clear in policy, implementation is far more complex. Defining concepts like "excessive use" or "extreme emotions" in a practical, enforceable way will challenge both regulators and firms.

Systems that are too sensitive may irritate users engaged in harmless, extended chats. Settings that are too permissive could allow at-risk users to go unnoticed. The mandatory two-hour reminder is a straightforward tool, but the obligation to act "when signs can be detected" leaves considerable room for interpretation.

Developers of AI safety measures have faced similar hurdles. Content filters are often imprecise, and adding psychological monitoring introduces new layers of complexity. Assessing whether a user is developing an unhealthy bond requires inferring mental state from text—a capability current AI lacks reliability in.

The draft is now open for public feedback, with final regulations anticipated in 2026. If enacted as proposed, China would establish the world's most detailed regulatory framework for AI companion products.

A Global Reckoning

Concurrent regulatory steps in China and California signal that worries about AI companion addiction have become a pressing global issue, transcending political systems.

For AI companies, the writing is on the wall: the unregulated phase for AI companions is closing. Whether through Chinese administrative rules, California's civil lawsuits, or future U.S. federal laws, platforms will be required to shield users from potential harm caused by their own products.

The central question is no longer if regulation will arrive, but whether the proposed safeguards will be effective. China's method of mandated monitoring and intervention may face significant practical hurdles.

What remains evident is that the AI companion market has become too vast and impactful for governments to overlook. The chatbots people form emotional attachments with are no longer novelties—they are products used by hundreds of millions, with confirmed instances of serious harm. Regulation, however imperfect, was unavoidable. The discussion now centers on whether these specific rules can safeguard vulnerable users without hindering a technology that many find genuinely beneficial.