According to Forbes, China has posted a newly drafted set of AI laws for public comment, with a specific focus on regulating AI services that impact human mental health. The draft, titled “Interim Measures for the Administration of Artificial Intelligence Anthropomorphic Interactive Services,” was published online on December 27, 2025, and the public feedback period runs until January 25, 2026. The proposed rules have a broad jurisdictional scope, applying to any AI service accessible to the public within China, even if the technology is based overseas. This comes as the United States has no overarching federal law governing AI in mental health, with only a handful of states like Illinois, Utah, and Nevada enacting their own limited regulations. The draft law includes provisions aimed at preventing AI from providing “false promises that seriously affect user behavior” or services that “damage social interpersonal relationships.”
The Global Regulatory Vacuum
Here’s the thing: the world is scrambling to figure out how to handle AI in sensitive areas like mental health, and nobody has really nailed it yet. In the US, we’ve got a patchwork. A few states have laws, but they’re all different. Congress talks a big game but hasn’t passed anything. Meanwhile, millions of people are already using ChatGPT and other chatbots as de facto therapists because it’s cheap, private, and available 24/7. That’s a massive, uncontrolled experiment happening in real-time. So when a major power like China steps in with a draft law, everyone pays attention. It’s not just about their domestic policy; it sets a potential template and creates pressure for other nations to act.
rules”>Decoding The Draft Rules
The Forbes analysis, using a Google Translate version of the document, highlights two critical articles. Article 2 defines the scope, and it’s broad: it covers AI that simulates human personality and interacts emotionally. That’s basically every modern LLM when used for companionship or advice. More crucially, the jurisdiction covers any service accessible in China. That’s a shot across the bow to global AI makers—if your chatbot is available in China, you play by these rules. Then there’s Article 7. The provision about “false promises” and damaging relationships is fascinatingly vague. What constitutes a “serious” effect on behavior? Persuading someone to leave a job? Or to end a friendship? The ambiguity is either a flaw or a feature, giving regulators wide latitude. But it shows they’re thinking about the insidious, long-term psychological impacts, not just immediate safety.
The Core Challenge: Defining The Undefinable
This gets to the heart of why regulating this stuff is so hard. How do you write a law for something as nuanced and personal as mental health advice? A human therapist can have a bad day, give questionable advice, and it’s a malpractice issue. An AI system can hallucinate, reinforce delusions, or give dangerously generic advice to millions simultaneously. The Chinese draft seems to be trying to set a perimeter fence, saying AI shouldn’t cross certain lines. But enforcing that is a nightmare. Does monitoring for compliance require constant, intrusive scrutiny of AI-human conversations? And who decides what damages a social relationship? It’s a classic tech regulation problem: you either write rules so broad they’re unworkable, or so narrow they’re useless by launch day.
What It Means For Everyone Else
Look, I don’t think the US will copy China’s approach verbatim. The contexts are too different. But this draft does two important things. First, it legitimizes mental health as a critical, separate category for AI regulation—it’s not just another app. Second, it reinforces that geographic borders are porous online. If you’re a US AI company with global aspirations, you now have another major regulatory regime to consider. The pressure for some kind of coherent US federal framework just went up a notch. Basically, the world is conducting a bunch of parallel experiments in AI governance right now. China’s is one of the more assertive. We’ll see which approaches stick, and which ones collapse under their own weight or get watered down by industry lobbying. The race to regulate the mind of the machine is officially on.
