A student walks into a translucent pod between classes, closes the door, and starts talking to something that won't grade them, judge them, or call their parents. Across China, that scene is moving from pilot to procurement. The catch, repeated by every vendor and teacher involved, is that nobody claims the machine can replace a counselor.
From pilots to procurement
The early movers were universities. At Central University of Finance and Economics (中央财经大学), the "AI Listening Mind Cabin (AI聆心小屋)" opened in November 2025 and was upgraded in April 2026. Built with Tsinghua University's Interactive AI research group (清华大学交互式人工智能课题组) and its spin-off, it pairs a large model with a private physical space. In a study run there, 92.86% of participating students said the AI gave useful advice, and the team estimates the current model performs at the level of a novice counselor with about 50 hours of practice — good for everyday stress, not for deep trouble.
The shift now is to K–12 scale. Anzhi Technology (安智科技), with Tsinghua Zhipu · Lingxin Intelligence (清华智谱·聆心智能), has deployed its "Xiaozhi Listening Mind AI Space Station (小智聆心AI空间站)" in 300+ schools, including Tsinghua High School (清华附中) and Peking University High School (北大附中), across 28 provinces, serving over 6 million teachers and students. The engine is Emohaa 2.5 (Emohaa2.5), a campus-specific mental-health model developed over roughly a decade. The "space station" ships in four forms:
- Pod version (太空舱版): a standalone cabin for private, anonymous talking.
- Classroom version (教室版): embedded in normal rooms for mental-health lessons.
- Corridor version (走廊版): a self-service decompression terminal at break time.
- Calm-corner version (减压角版): placed in libraries and activity areas.
Other districts went their own way. Beijing Hongzhi Middle School (北京宏志中学) installed 50 AI screening devices; iFlytek's (科大讯飞) "AI Stress-Relief Planet (AI减压星球)" runs in soundproof glass rooms; at Changping No.1 Middle School (昌平一中), the "Zizai (自在)" counseling model launched in May 2025 and, over two months, logged 1,300+ sessions, hit a 45% family weekly-active rate, and cut high-risk screening time from two weeks to three days. In Xiamen Experimental Primary School (厦门实验小学), a virtual counselor named "Duoduo (朵朵)" reads facial and eye-movement cues on camera to catch a child's hesitation.
Why now
Three forces line up at once:
- Demand is documented. The 2025 Mental Health Blue Book (心理健康蓝皮书) from the Institute of Psychology at the Chinese Academy of Sciences puts the adolescent depression detection rate at 24.6%.
- Supply is thin. Industry estimates put the shortage of full-time rural school counselors at around 68%, and most schools cannot hire their way out quickly.
- Policy is pushing. In October 2025, the Ministry of Education's office issued the Ten Measures to Strengthen Student Mental Health (进一步加强中小学生心理健康工作十条措施). In the fall 2026 term, education bureaus in Wuhan, Liuzhou, and Jincheng told schools to finish a full mental-health assessment within one month of opening.
Money is following. Industry analyses estimate China's education mental-health AI application market passed 12 billion RMB (≈ US$1.7B / HK$13B) in 2025, as schools fold AI devices into their infrastructure budgets.
Trend reading
According to Li An, chief scientist at BrainNet (脑机网), the campus deployments show a useful pattern: AI works best as an always-available first listener that widens the safety net, while the licensed counselor stays the person who closes the loop. The risk is treating the cabin as a substitute rather than a triage front door.
Honest limitations
- The model is junior. At a "50-hour novice" level, it handles structured methods like CBT and mood tracking, but misses the deeper work — personality growth, family dynamics — that a trained counselor probes.
- It is blind to silence. Project leads admit the AI struggles to read non-verbal cues; a child quietly crying may not register, and there is no real human "presence."
- Quality varies. Educators warn some models give clumsy or even counterproductive replies, and vendor numbers (300 schools, 6 million users) are company-reported, not independently audited.
- Privacy is unresolved. These systems collect voice, text, and sometimes video from minors; schools must clear compliance, data-security, and teaching-fit reviews, and not every vendor meets the bar.
- Not diagnosis. None of the deployments claim to diagnose or treat; the final warning call rests with a teacher.
What readers can do now
- Students and parents: treat an AI station as a first-aid ear, not therapy. Use it to name a feeling, then escalate to a real counselor when something feels heavy.
- Schools: keep the teacher as the final decision-maker on any warning, and pair the tech with hiring or training human counselors rather than replacing them.
- Policymakers: fund the human side of the equation and set clear data-security and age-appropriate standards before scaling hardware into classrooms.