other · Frontiers in psychology · la publicación, 14 oct 2026 · gratis
Una universidad china ya empezó a medir la salud mental de sus estudiantes; las ocho terapias que planean probar siguen sin arrancar
El plan cubre a unos 42,000 estudiantes en siete universidades y no reporta ningún resultado: no muestra que la inteligencia artificial detecte mejor la depresión ni que ninguna de las terapias funcione.
Versión breve · la versión detallada sigue, unos 8 min
Pregunte al chat de weeklyAI
Pregúnteme sobre la IA en su vida: le señalo las notas que weeklyAI ha publicado sobre el tema, esta semana y antes.
Las conversaciones se guardan mientras exista weeklyAI, para mejorar la publicación. Se responde en el idioma en que usted escribe.
- El estudio, de un vistazo
- Quiénes
- Estudiantes universitarios de 18 a 30 años en China
- Cuántos
- unos 42,000 estudiantes en siete universidades
- Dónde
- China, siete provincias
- Cuándo
- el cohorte comenzó en septiembre de 2025; el artículo se publicó el 14 de octubre de 2026
- Tipo de estudio
- un plan de investigación (protocolo) sin resultados
- Quién lo hizo
- Universidad Médica de Henan y seis universidades chinas más
- El límite que importa
- No reporta ningún resultado: ni la inteligencia artificial ni las ocho terapias han demostrado nada todavía.
En abril de 2025, un comité de ética de la Universidad Médica de Henan aprobó el primer reclutamiento. En septiembre de 2025 empezaron a inscribir estudiantes en esa universidad, en el centro de China.
El plan es amplio. El plan es inscribir a unos 42,000 estudiantes en siete universidades repartidas por siete provincias chinas. Los que entren deben tener entre 18 y 30 años y estar inscritos en una de esas universidades. Cada semestre, durante cinco años, contestarían varios cuestionarios sobre ánimo, ansiedad, sueño, estrés e ideas suicidas. Son nueve rondas después de la primera.
Un grupo más pequeño, de hasta 400 estudiantes por centro cada año, daría además datos del cuerpo y del cerebro: electroencefalograma, resonancia magnética, oxigenación cerebral, ritmo del corazón, seguimiento de ojos, grabaciones de voz y de rostro. Hay un componente de muestras biológicas descrito, pero todavía no está activado.
Ocho ensayos separados están planeados, cada uno con dos grupos y con sus propios criterios, su comparación y su medida principal. Terapia cognitivo-conductual, terapia interpersonal y de ritmos sociales, ejercicio físico, música, realidad virtual, arte, biofeedback y una intervención de regulación del estrés por los cinco sentidos. Todos siguen en la etapa de planeación: no han reclutado ni asignado a nadie.
El artículo lo dice con claridad. Los modelos de inteligencia artificial se desarrollarán y evaluarán fuera de línea. No decidirán diagnósticos, ni cómo manejar el riesgo de suicidio, ni quién entra a un ensayo, ni qué tratamiento recibe una persona. Cualquier respuesta positiva en el cuestionario de ideas suicidas activa una revisión humana, no un programa.
No se reporta ningún resultado. Los autores dicen que queda por demostrar si este tipo de datos mejora la detección, si los modelos sirven fuera de estas universidades y si el programa tiene utilidad clínica.
Las siete universidades se eligieron por colaboraciones de investigación, no al azar. La muestra es mayoritariamente de etnia han. Los estudiantes con un diagnóstico previo de depresión o ansiedad generalizada quedan fuera.
Nada de esto dice qué permite la ley en México, Colombia, Brasil, Estados Unidos ni Canadá sobre el uso de su rostro, su voz o sus datos de salud. Es un plan de investigación en China, y no ha producido todavía ni un dato que alguien pueda usar para exigir algo a una autoridad.
Qué significa para usted
Ese plan chino no cambia nada de lo que usted puede exigir hoy en México, Colombia, Brasil, Estados Unidos ni Canadá sobre su rostro, su voz o sus datos de salud. Si le preocupa el uso de la inteligencia artificial en la escuela o en el trabajo, la pregunta útil es qué ley de su país lo cubre y qué autoridad la vigila. Ningún resultado de este estudio sirve todavía para reclamar nada.
Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887
Quién pagó: El trabajo recibió apoyo de financiadores chinos, entre ellos la Fundación Nacional de Ciencias Naturales de China, varios programas provinciales de Henan y un fondo conjunto de ciencia y tecnología de Henan; el artículo afirma que los financiadores no tuvieron papel alguno en el diseño del estudio, la recogida de datos, el análisis, la interpretación ni la preparación del manuscrito, y no se acredita a ningún proveedor de equipos o programas informáticos.
No tome esto como consejo médico profesional.
Versión detalladaLos pasajes copiados del artículo, las ilustraciones y cada fuente con cuánto leímos de ella · 1589 palabras · unos 8 minLeerla →Cerrar
Un plan de cinco años para leer la mente del estudiante, y ocho tratamientos que aún no empiezan
Siete universidades chinas quieren seguir a 42,000 alumnos con cuestionarios, voz, rostro, imágenes cerebrales y muestras biológicas. El artículo no reporta ni un solo resultado.
Siete universidades de China, repartidas por siete provincias, se proponen acompañar a unos 42,000 estudiantes de entre 18 y 30 años durante cinco años, con una evaluación cada semestre. El programa se llama CAMPUS y el artículo que lo describe se publicó el 14 de octubre de 2026. No presenta hallazgos: lo dice con todas sus letras, este texto describe el protocolo y no reporta datos de participantes ni resultados del cohorte1. Así que lo que sigue no es el descubrimiento de nada. Es el plano de una obra que apenas comenzó.
La pieza central es una batería de medición. Los alumnos contestan cuestionarios de depresión, ansiedad, insomnio, estrés, ideación suicida, anhedonia y adicción al celular. Además se les toman electroencefalogramas, resonancias magnéticas, mediciones de oxigenación cerebral, de variabilidad del ritmo cardiaco y de movimiento ocular; se graba su voz y se analiza su expresión facial2. Sobre todo eso se planea construir un modelo de inteligencia artificial que combine las distintas fuentes de datos y señale qué tanto pesa cada una en la predicción3. La palabra que el propio protocolo usa para describir esas tareas es "cuestionario": clasificar a alguien como posible caso según sus respuestas, no diagnosticarlo4.
Conviene detenerse en ese punto, porque es donde suele perderse la cabeza. Una revisión incluyó 103 fuentes, de ellas 95 estudios primarios y 8 síntesis de evidencia; en 83 estudios se reportó desempeño predictivo o diagnóstico, pero solo en uno se reportó efecto sobre los síntomas, y ningún estudio primario ofreció una evaluación prospectiva dedicada a la seguridad de la herramienta56. Del resumen de esa revisión pudimos leer solo el resumen; el texto completo está detrás de una suscripción56. Otra revisión, centrada en chatbots de salud mental, encontró que las fallas documentadas incluían no detectar la ideación suicida y entregar información clínica inexacta; también de esta leímos únicamente el resumen7. El contraste con CAMPUS es claro y hay que decirlo sin adornos: aquí no hay chatbot que converse con nadie, y los modelos, según el protocolo, se desarrollan y evalúan fuera de línea, sin decidir diagnóstico, disposición de riesgo, elegibilidad ni asignación de tratamiento8.
Piense en lo que cubre una consulta de rutina: le toman la presión, le preguntan cómo durmió, le revisan la garganta. Aquí el equivalente es un cuestionario, una grabación de voz, una resonancia y, en el futuro, un análisis de sangre. Lo curioso es que el propio protocolo reconoce que nada de eso se ha activado todavía en su parte biológica: sangre, orina, heces y saliva son candidatos, pero el calendario y los procedimientos se definirán antes de empezar, y no se recolectará ninguna muestra hasta que estén aprobados los manuales, los consentimientos y las enmiendas éticas910. La pregunta que salta es obvia: ¿qué se le pide al estudiante que firme hoy y qué queda para después? La respuesta está en el texto: la voz, el video del rostro, las pruebas fisiológicas, la resonancia y las muestras futuras tienen consentimiento aparte, componente por componente11.
Ahí aparece el primer gesto razonable del diseño. Quien no quiera que le graben la cara puede seguir en el cohorte; negarse a una prueba no lo saca del estudio1213. El acceso a datos identificables, biometría cruda, material genético y la llave que une el código con el nombre queda limitado al personal autorizado y bajo supervisión institucional14. Las grabaciones de voz, los videos faciales, las imágenes y los datos genómicos no se depositarán en repositorios públicos15. Y si alguien de fuera pide acceso a muestras o a datos sensibles, tendrá que pasar por revisión científica e institucional16. Son candados. La pregunta es quién tiene la llave y qué pasa cuando se pierde.
El tamaño del aparato merece una escala. Cuarenta y dos mil estudiantes es más o menos la población de una ciudad mediana de América Latina llenando un formulario cada seis meses durante cinco años. El protocolo calcula que, con una deserción anual del 15 por ciento, quedarían unos 18,600 al final del periodo. Son cifras de planeación, no resultados. La parte más pesada, la de resonancias y electroencefalogramas, se limitaría a unas 400 personas por centro por año, y el análisis molecular arrancaría con unos 3,000 participantes. Nada de esto se ha medido aún.
Las universidades no se eligieron al azar de una lista nacional: se eligieron por colaboraciones de investigación ya existentes, y el propio protocolo admite que el cohorte no representa al país ni a todos los estudiantes chinos; la muestra es mayoritariamente de etnia han. También admite que las pruebas más caras podrían atraer a un subconjunto distinto de alumnos, que los aparatos varían de un centro a otro y que se excluye a quienes ya tienen diagnóstico de depresión o ansiedad, lo que deja fuera buena parte del espectro grave. Son límites que el documento declara, y hay que reconocérselos.
Sobre los tratamientos, el dato que ordena todo: los ocho ensayos aleatorizados siguen en etapa de planeación. No hay reclutamiento, ni asignación al azar, ni sesiones impartidas. El protocolo describe terapias de conducta, de ritmos sociales, música, ejercicio, realidad virtual, arte, biofeedback y una intervención de cinco sentidos, con comparadores que se definirán antes de activar cada ensayo. Cada universidad nueva deberá obtener aprobación de su comité de ética antes de tocar a un solo estudiante17. Es decir: por ahora no hay nada que decir sobre si alguna de estas terapias funciona en este programa, porque nadie ha recibido ninguna.
Así lo leemos nosotros. El patrón que se repite en estos sistemas es que casi nunca se discute en público cómo se entrenó la herramienta ni quién responde cuando se equivoca, y el peso del error cae de manera desigual sobre quienes ya cargan desventajas. Si algún día estos modelos de riesgo se usaran para decidir quién recibe atención, los estudiantes con menos recursos podrían pagar los errores sin que nadie rinda cuentas ante ellos. Nos haría cambiar de idea ver auditorías independientes publicadas, explicaciones claras del entrenamiento y evidencia de que los errores se reparten por igual entre todos los grupos.
La segunda cosa que vemos es el desequilibrio de siempre: quien ordena los datos manda más que quien los entrega. Aquí se recogerán cuestionarios, voz, rostro, imágenes del cerebro y muestras biológicas de decenas de miles de personas, y el control quedará del lado de la universidad y del centro coordinador. Lo que el estudiante necesita saber es qué se guarda, dónde, por cuánto tiempo y cómo pedir que se borre. Cuando una institución o una empresa use inteligencia artificial con sus datos, pregunte quién entrenó el sistema, con qué datos y a quién puede reclamar si le perjudica. En su país, averigüe si existe una autoridad de protección de datos a la que dirigir esa pregunta por escrito. Antes de firmar cualquier consentimiento, pida por escrito qué datos se guardan, quién los ve y cómo puede retirarlos, y guarde copia.
Hay una tercera cosa que vale la pena vigilar, aunque este plan no la provoque. Un sistema que marca a alguien como "en riesgo" y que solo un programa decide qué sigue puede llevar a que el personal aplique etiquetas sin cuestionarlas. Lo contrario se vería si quien recibe las alertas tiene formación para dudar, puede corregir la clasificación y existen canales para que el estudiante pida una segunda opinión humana. Si una institución lo clasifica con una herramienta automática, pregunte quién revisa esa clasificación. En la escuela o el trabajo, pida que las decisiones que le afectan no queden solo en manos de un programa.
Falta el dato que más pesa para quien tiene un hijo, un sobrino o un amigo en la universidad: el protocolo dice que cualquier respuesta positiva en el cuestionario de ideación suicida dispara una alerta electrónica revisada por una persona, y que los modelos de inteligencia artificial no se usarán para manejar riesgos en tiempo real ni sustituirán ni retrasarán la valoración profesional18. El documento no reporta casos, no cuenta historias y no ofrece cifras de intentos ni de muertes. Se sabe que hay muertes relacionadas con estos padecimientos, y ahí queda.
Nos queda una duda honesta sobre el calendario: el cohorte ya comenzó en septiembre de 2025 en una universidad, y el protocolo maestro lleva fecha de septiembre de 2026, revisado durante la evaluación de la revista. Los autores afirman que quienes desarrollaron los planes de análisis no habían visto datos de resultados antes de esa fecha. No tenemos manera de verificarlo por nuestra cuenta; lo tomamos como está escrito.
Lo que este documento permite, y es bastante, es saber qué preguntar. Si en su país se anuncia un programa parecido, con inteligencia artificial para detectar alumnos en riesgo, ya tiene el guion: pida el protocolo, pregunte qué mide, quién entrena el modelo, qué consentimiento se firma por separado para la voz y el rostro, quién responde por un error y cómo se pide una revisión humana. Aquí, en el plano chino, esas respuestas están escritas. Allá, habrá que exigirlas.
Volvamos al formulario que ese estudiante llena cada semestre. Ahora usted sabe que detrás hay un modelo que aún no existe, ocho tratamientos que aún no empiezan y una promesa de candados. La pregunta que puede llevar a la siguiente junta escolar o universitaria es sencilla: ¿quién firma cuando el sistema se equivoca?
De dónde sale cada dato de contexto, y cuánto leímos de cada documento
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - el artículo del que trata esta nota — el artículo completo — el pasaje: "This article describes the study protocol and does not report participant-level or aggregate findings from the earlier developmental activities or the longitudinal cohort."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Multidimensional screening integrates psychological self-report (PHQ-9, GAD-7, ISI, BSI-CV, and others), physiological measurements (EEG, MRI, fNIRS, HRV, eye tracking, facial-expression analysis, ERP, voice acoustics), and an exploratory biospecimen component for multi-omics profiling"
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - el artículo del que trata esta nota — el artículo completo — el pasaje: "A Transformer-based multimodal AI framework with SHAP-based feature attribution will be developed and internally evaluated for questionnaire-defined risk stratification and exploratory treatment-response prediction."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The AI analyses will address questionnaire-defined screening and prediction tasks rather than clinical diagnosis. No structured diagnostic interview or independent clinical adjudication is planned."
- Dhamayanti M, Susaldi S, Hanifa F, Agustini Y, Emilia O, Hardinsyah H, et al. (2026). Artificial intelligence-enabled mental health applications for adolescents: a scoping review of safety, effectiveness and clinical implementation. BMJ Open. 10.1136/bmjopen-2026-125586 — solo el resumen - el texto completo está tras una suscripción — el pasaje: "Predictive or diagnostic performance was reported in 83 studies (87.4%), but symptom-level effectiveness in only 1 (1.1%)."
- Dhamayanti M, Susaldi S, Hanifa F, Agustini Y, Emilia O, Hardinsyah H, et al. (2026). Artificial intelligence-enabled mental health applications for adolescents: a scoping review of safety, effectiveness and clinical implementation. BMJ Open. 10.1136/bmjopen-2026-125586 — solo el resumen - el texto completo está tras una suscripción — el pasaje: "No primary study provided a dedicated prospective AI-safety evaluation."
- Olisaeloka L, Richardson CG, Wang AY, Munthali RJ, Vigo DV. (2026). Safety Mechanisms and Risk Mitigation in Generative AI Mental Health Chatbots: A Systematic Scoping Review. Healthcare. 10.3390/healthcare14101395 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Documented safety failures included missed suicidal ideation and provision of inaccurate clinical information."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - el artículo del que trata esta nota — el artículo completo — el pasaje: "All models will be developed and evaluated offline. Their outputs will not determine clinical diagnosis, suicide-risk disposition, trial eligibility, randomized allocation, treatment selection, intervention intensity, or protocol modification."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The biospecimen component has not yet been activated. Blood, urine, stool, and saliva are candidate biological matrices, but the collection schedule and matrix-specific procedures will be finalized before activation."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - el artículo del que trata esta nota — el artículo completo — el pasaje: "No biospecimen collection will begin until the relevant standard operating procedures, consent materials, and required ethics amendments have been approved."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Voice recording, facial-expression video recording, physiological and neuro-functional assessments, MRI, and future biospecimen collection for genetic and other multi-omics analyses will be subject to separate component-specific consent or consent options"
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Participants may decline or discontinue any optional component without affecting their continued participation in the core cohort."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Inability or unwillingness to complete an individual optional component, including voice or facial-expression recording, EEG, ERP, HRV, eye tracking, fNIRS, MRI, or biospecimen collection, will not exclude the participant from the longitudinal cohort."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Access to identifiable information, raw biometric data, genomic data, biospecimens, and the linkage key will be limited to authorized personnel according to their study roles and will be subject to institutional oversight."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Raw voice recordings, facial-expression videos, imaging data, genomic data, and other data carrying a material risk of re-identification will not be deposited in public repositories."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Requests for access to biospecimens or sensitive individual-level data will require scientific and institutional review and must be consistent with the approved consent provisions, ethics approvals, and applicable data-use agreements."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Each additional participating university will obtain approval from its responsible ethics committee before site activation, participant enrolment, or data collection."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - el artículo del que trata esta nota — el artículo completo — el pasaje: "AI models will be developed and evaluated offline; their outputs will not be used for real-time safety management, clinical risk assignment, or disposition and will not replace or delay professional assessment or emergency action."
Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887
Quién pagó: El trabajo recibió apoyo de financiadores chinos, entre ellos la Fundación Nacional de Ciencias Naturales de China, varios programas provinciales de Henan y un fondo conjunto de ciencia y tecnología de Henan; el artículo afirma que los financiadores no tuvieron papel alguno en el diseño del estudio, la recogida de datos, el análisis, la interpretación ni la preparación del manuscrito, y no se acredita a ningún proveedor de equipos o programas informáticos.
No tome esto como consejo médico profesional.
other · Frontiers in psychology · the paper, 14 Oct 2026 · free
China Is Planning a Five-Year Mental Health Study Across Seven Universities. No Results Exist Yet.
A plan describes screening 42,000 students with questionnaires, brain scans and voice recordings. It reports nothing yet about whether any of it helps anyone.
Short version · the longer version follows, about 8 min
Ask the weeklyAI chat
Ask me about AI in your life: I point you to the stories weeklyAI has published on it, this week and before.
Conversations are saved for as long as weeklyAI exists, to improve the publication. Answers come in the language you write in.
- The study at a glance
- Who
- University students in China
- How many
- about 42,000 students
- Where
- seven universities in China
- When
- enrolment began September 2025
- Kind of study
- survey
- Who did it
- Mental Health and Artificial Intelligence Research Center, Henan Medical University, and others
- The limit that matters
- It is a research plan; no results are reported yet
The AI models are built and tested offline only.
In April 2025, an ethics committee at Henan Medical University in China approved the first student recruitment and mental health assessment there. Enrolment of students there began in September 2025.
The plan behind it is called CAMPUS. It aims to enrol about 42,000 students, roughly 6,000 from each of seven Chinese universities, and to assess them once a semester for five years. The questionnaires ask about mood, anxiety, sleep, stress and suicidal thoughts. The universities sit in different regions of China and were chosen through existing research partnerships.
A smaller group, up to 14,000 students over the five years, would also give physical measurements. These include brain wave recordings, brain scans, heart-rate readings, eye tracking, voice recordings and video of facial expressions. The plan also describes collecting blood, urine, stool and saliva for molecular analysis. That part has not been switched on. It needs its own procedures and ethics approval first, and students must agree to it separately.
Eight separate randomized trials are planned. Each would test one intervention, such as cognitive behavioral therapy, music, exercise, virtual reality, art, biofeedback or a five-sensory stress program, against its own comparison group. Each would aim to enrol 420 students, a figure that allows for some students dropping out. All eight are still in planning. None has begun, and no student has been assigned to a treatment.
The article is clear about the software. The AI models are built and tested offline only. Their output will not decide a diagnosis, how a suicidal student is handled, who enters a trial, or which treatment anyone receives. A positive answer on the suicide questionnaire triggers a human safety review, whatever the software says.
No results are reported. The authors say whether these tools travel well to other settings, whether they are useful in practice, and whether the extra measurements improve screening all remain to be shown.
The study is set in China and says nothing about the law in Latin America, the United States or Canada. It cannot tell you what any government or court where you live permits with your face, voice or health data, or what you can demand. It is a research plan, not a ruling. Whether the plan works is still unknown.
What this means for you
For now, nothing in this plan changes what a company may do with your face or voice where you live, and no agency or court has ruled on it. What you can watch is whether the Chinese universities ever publish results, and whether any government near you cites such a plan as a reason to loosen or tighten its own rules on health data.
Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887
Who paid: The work was supported by Chinese funders including the National Natural Science Foundation of China, several Henan provincial programs and a Henan science and technology joint fund; the article states the funders had no role in study design, data collection, analysis, interpretation or manuscript preparation, and no equipment or software providers are credited.
Do not take this as professional medical advice.
The longer versionThe passages copied from the paper, the pictures, and every source with how much of it we read · 1605 words · about 8 minRead it →Close
China Plans a Five-Year AI Mental-Health Screening Program for 42,000 Students. No Results Yet.
Seven universities will collect faces, voices, brain scans and blood from students to train models. The paper is a plan, not a finding.
Consider what a student at one of seven Chinese universities might be asked to do. First, a set of questionnaires about mood, sleep, worry and stress, filled in on a phone app. Then, for some, a session in a lab: a cap of electrodes on the scalp, a scan in a magnetic resonance machine, a camera recording the face while the student answers open questions, a microphone recording the voice reading a passage, an eye tracker following the gaze. Later, perhaps, a blood sample, urine, stool, saliva.
This is the CAMPUS program, described in a protocol published on 14 October 2026. It is a five-year study across seven universities in seven provinces of China, aiming to enroll about 42,000 students aged 18 to 30, with roughly 6,000 from each campus. The authors are a large team from many institutions, including the Mental Health and Artificial Intelligence Research Center at the Second Affiliated Hospital of Henan Medical University, the other six universities and Yale. The work is paid for by the National Natural Science Foundation of China and several Henan provincial programs; the authors declare no commercial conflicts.
The paper reports no results. It says so itself: "This article describes the study protocol and does not report participant-level or aggregate findings"1. No student has yet been screened by the AI, no trial has randomized anyone, no biological sample has been collected. What exists is a design: a cohort followed once a semester for five years, eight small randomized trials still in planning, and an exploratory set of AI and molecular analyses. The eight trials, each planned at 420 participants, cover cognitive behavioral therapy, interpersonal and social rhythm therapy, physical exercise, music, biofeedback, virtual reality, expressive art and a "five-sensory" stress program. None has begun.
The technology is worth picturing before the word "AI" does any work. The plan is to feed questionnaire answers, brain signals, heart-rate variability, eye movements, facial expressions and voice acoustics into a single model, built on a "Transformer" architecture, the same family of design behind large language models23. Each type of data is converted into a compact numerical summary, and the summaries are weighed against one another to produce a risk category or a prediction3. The researchers will also use a method called SHAP to show which inputs pushed a given prediction up or down3. The target is not a diagnosis: the paper states the models address "questionnaire-defined screening and prediction tasks rather than clinical diagnosis," and no structured diagnostic interview is planned4.
Now the limit that matters most, in the authors' own words. All models "will be developed and evaluated offline," and their outputs "will not determine clinical diagnosis, suicide-risk disposition, trial eligibility, randomized allocation, treatment selection, intervention intensity, or protocol modification"5. For suicide risk specifically, the models will not be used "for real-time safety management, clinical risk assignment, or disposition" and will not "replace or delay professional assessment or emergency action"6. A positive answer on the suicide-ideation questionnaire triggers a human response pathway with three levels, from a call within 24 hours to emergency services. The algorithm stays out of it. The paper describes the questionnaire as an initial safety screen that does not determine risk level or what happens next; that is decided by a qualified mental health professional.
Compare this with what the wider literature has found. A review of generative AI mental-health chatbots covered 21 studies in 11 countries7 and found that documented safety failures included missed suicidal ideation and inaccurate clinical information8. Most of those chatbots had at least one technical safeguard7, but human oversight was limited and crisis referral protocols were mostly underdeveloped910. A second review looked at 103 sources on AI mental-health tools for adolescents11 and found that predictive or diagnostic performance was reported in 83 studies, while symptom-level effectiveness appeared in just one12. No primary study in that review provided a dedicated prospective safety evaluation13. CAMPUS is, in part, an answer to that gap: a plan to test prospectively what others have only predicted.
The data-handling rules in the protocol are unusually explicit, and the reader should see them. Voice recordings, facial videos, imaging and genomic data "will not be deposited in public repositories"14. Trained model weights and embeddings will undergo a disclosure-risk review before sharing, because they may retain information about the people they were trained on15. Each intrusive component — the camera, the scanner, the blood draw — requires its own separate consent16. A student may decline any optional piece without losing their place in the main cohort1718. Access to raw biometric and genomic data is limited to authorized personnel under institutional oversight19, and outside requests for samples or sensitive data require scientific and institutional review20. The biological component has not been activated; blood, urine, stool and saliva are candidates, and nothing will be collected until procedures, consent materials and ethics amendments are approved2122. Each new university must obtain local ethics approval before it starts23.
Here is how we read it. A program that gathers faces, voices, brain scans and blood from tens of thousands of students is building a shared technical resource, and the people whose bodies supply it will have limited practical say over what happens next. The consent forms exist; the question is whether a student can later withdraw one recording, one sample, one scan, and whether a refusal is honored without quiet penalty. We would be wrong if the program published, in plain language, exactly what each student's face and sample may and may not be used for, and showed that refusals cost nothing. Until then, when a school or clinic asks for your face, your voice or a sample "for research," ask in writing which specific uses are covered, whether you can withdraw a single item later, and who decides new uses. Keep the answer.
The second thing we notice is the asymmetry of information. A student screened by a model will rarely learn what the model concluded, on what basis, or who inside the university sees the result. The protocol is careful to say the AI will not decide diagnosis, allocation or eligibility5. It does not say the student will be told their own score. That gap is where trouble usually starts: a number circulates, a counselor acts on it, and the person it describes never sees it. You can watch for this where you live by asking your university who receives screening results and whether those results can affect your standing as a student. Ask for your own result and the reason for it, in writing. A questionnaire score is a question to discuss with a qualified person, not a verdict.
The study's own limitations are stated plainly. The seven universities were chosen through existing research collaborations, not by random sampling from a national frame, so the cohort is not nationally representative, and the sample is predominantly Han Chinese1. Five years of follow-up will lose people to attrition. The heaviest components — MRI, blood, stool — may attract a self-selected group, which can bias the biological subsample. Devices differ by center, and the paper acknowledges that residual differences may affect pooled estimates. Students with a prior diagnosis of major depression or generalized anxiety are excluded, which limits the range of severity studied. And algorithmic bias — different model performance across sex, academic stage or center — is named as a recognized risk. The validation planned is internal only: held-out participants, later time periods, and leaving one university out at a time. That is not independent external validation, and the paper says so.
What would have to happen before any of this means anything for a student or a family? The cohort must actually enroll and retain its participants across nine semester waves. The biological component must be activated with approved procedures. Each of the eight trials must finalize its comparator, manual, statistical plan and registration before the first randomization. And the models must be tested against simpler benchmarks to see whether combining faces, voices and brain signals actually improves prediction over a plain questionnaire. The paper says the combined models will be compared with simpler models that use one type of data or standard statistical methods. It does not report the answer.
We also read the protocol's own honesty as a warning against overreading it. The authors say the AI will not decide who gets help5. That is a promise about this program, in this place, at this stage. It is not a statement about what other institutions, vendors or governments might do with similar tools. The pattern to watch is familiar: a screening system arrives labeled as support, then quietly becomes a gate. You can tell whether that is happening by one question — who sees the output, and what can they do with it? If the answer is "only the research team, and nothing," the promise holds. If the answer is "the dean's office," ask why.
The questionnaires, the electrode cap, the camera and the blood draw are all in service of a single hope: that a machine might notice distress earlier than a busy counselor can. That hope is reasonable and untested. The plan is careful, the safeguards are written down, and the results do not exist yet. When your own school or clinic proposes something similar, the useful question is not whether the technology is impressive. It is: what will they tell me about what they concluded, and who else will they tell?
Where each piece of context comes from, and how much of it we read
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - the article this story is about — the whole article — the passage: "This article describes the study protocol and does not report participant-level or aggregate findings from the earlier developmental activities or the longitudinal cohort."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - the article this story is about — the whole article — the passage: "Multidimensional screening integrates psychological self-report (PHQ-9, GAD-7, ISI, BSI-CV, and others), physiological measurements (EEG, MRI, fNIRS, HRV, eye tracking, facial-expression analysis, ERP, voice acoustics), and an exploratory biospecimen component for multi-omics profiling"
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - the article this story is about — the whole article — the passage: "A Transformer-based multimodal AI framework with SHAP-based feature attribution will be developed and internally evaluated for questionnaire-defined risk stratification and exploratory treatment-response prediction."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - the article this story is about — the whole article — the passage: "The AI analyses will address questionnaire-defined screening and prediction tasks rather than clinical diagnosis. No structured diagnostic interview or independent clinical adjudication is planned."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - the article this story is about — the whole article — the passage: "All models will be developed and evaluated offline. Their outputs will not determine clinical diagnosis, suicide-risk disposition, trial eligibility, randomized allocation, treatment selection, intervention intensity, or protocol modification."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - the article this story is about — the whole article — the passage: "AI models will be developed and evaluated offline; their outputs will not be used for real-time safety management, clinical risk assignment, or disposition and will not replace or delay professional assessment or emergency action."
- Olisaeloka L, Richardson CG, Wang AY, Munthali RJ, Vigo DV. (2026). Safety Mechanisms and Risk Mitigation in Generative AI Mental Health Chatbots: A Systematic Scoping Review. Healthcare. 10.3390/healthcare14101395 — only the abstract - the full text could not be fetched — the passage: "Twenty-one studies across 11 countries were included. Most interventions incorporated at least one technical safety mechanism, most commonly fine-tuning and prompt engineering."
- Olisaeloka L, Richardson CG, Wang AY, Munthali RJ, Vigo DV. (2026). Safety Mechanisms and Risk Mitigation in Generative AI Mental Health Chatbots: A Systematic Scoping Review. Healthcare. 10.3390/healthcare14101395 — only the abstract - the full text could not be fetched — the passage: "Documented safety failures included missed suicidal ideation and provision of inaccurate clinical information."
- Olisaeloka L, Richardson CG, Wang AY, Munthali RJ, Vigo DV. (2026). Safety Mechanisms and Risk Mitigation in Generative AI Mental Health Chatbots: A Systematic Scoping Review. Healthcare. 10.3390/healthcare14101395 — only the abstract - the full text could not be fetched — the passage: "During intervention delivery, detailed onboarding with role clarification was common, but human oversight was limited."
- Olisaeloka L, Richardson CG, Wang AY, Munthali RJ, Vigo DV. (2026). Safety Mechanisms and Risk Mitigation in Generative AI Mental Health Chatbots: A Systematic Scoping Review. Healthcare. 10.3390/healthcare14101395 — only the abstract - the full text could not be fetched — the passage: "Crisis referral protocols varied in rigour but were mostly underdeveloped, and systematic adverse event monitoring was sparse."
- Dhamayanti M, Susaldi S, Hanifa F, Agustini Y, Emilia O, Hardinsyah H, et al. (2026). Artificial intelligence-enabled mental health applications for adolescents: a scoping review of safety, effectiveness and clinical implementation. BMJ Open. 10.1136/bmjopen-2026-125586 — only the abstract - the full paper is behind a subscription — the passage: "The database searches identified 11 167 records, and 1 additional evidence synthesis was identified and assessed during revision; 103 sources were included, comprising 95 primary studies and 8 evidence syntheses."
- Dhamayanti M, Susaldi S, Hanifa F, Agustini Y, Emilia O, Hardinsyah H, et al. (2026). Artificial intelligence-enabled mental health applications for adolescents: a scoping review of safety, effectiveness and clinical implementation. BMJ Open. 10.1136/bmjopen-2026-125586 — only the abstract - the full paper is behind a subscription — the passage: "Predictive or diagnostic performance was reported in 83 studies (87.4%), but symptom-level effectiveness in only 1 (1.1%)."
- Dhamayanti M, Susaldi S, Hanifa F, Agustini Y, Emilia O, Hardinsyah H, et al. (2026). Artificial intelligence-enabled mental health applications for adolescents: a scoping review of safety, effectiveness and clinical implementation. BMJ Open. 10.1136/bmjopen-2026-125586 — only the abstract - the full paper is behind a subscription — the passage: "No primary study provided a dedicated prospective AI-safety evaluation."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - the article this story is about — the whole article — the passage: "Raw voice recordings, facial-expression videos, imaging data, genomic data, and other data carrying a material risk of re-identification will not be deposited in public repositories."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - the article this story is about — the whole article — the passage: "Derived features, trained model weights, embeddings, and other model artifacts will undergo disclosure-risk review before sharing because they may retain information about the training data."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - the article this story is about — the whole article — the passage: "Voice recording, facial-expression video recording, physiological and neuro-functional assessments, MRI, and future biospecimen collection for genetic and other multi-omics analyses will be subject to separate component-specific consent or consent options"
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - the article this story is about — the whole article — the passage: "Participants may decline or discontinue any optional component without affecting their continued participation in the core cohort."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - the article this story is about — the whole article — the passage: "Inability or unwillingness to complete an individual optional component, including voice or facial-expression recording, EEG, ERP, HRV, eye tracking, fNIRS, MRI, or biospecimen collection, will not exclude the participant from the longitudinal cohort."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - the article this story is about — the whole article — the passage: "Access to identifiable information, raw biometric data, genomic data, biospecimens, and the linkage key will be limited to authorized personnel according to their study roles and will be subject to institutional oversight."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - the article this story is about — the whole article — the passage: "Requests for access to biospecimens or sensitive individual-level data will require scientific and institutional review and must be consistent with the approved consent provisions, ethics approvals, and applicable data-use agreements."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - the article this story is about — the whole article — the passage: "The biospecimen component has not yet been activated. Blood, urine, stool, and saliva are candidate biological matrices, but the collection schedule and matrix-specific procedures will be finalized before activation."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - the article this story is about — the whole article — the passage: "No biospecimen collection will begin until the relevant standard operating procedures, consent materials, and required ethics amendments have been approved."
- Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887 - the article this story is about — the whole article — the passage: "Each additional participating university will obtain approval from its responsible ethics committee before site activation, participant enrolment, or data collection."
Yang, G., Liu, R., Wang, C. et al. (2026). Comprehensive artificial intelligence-assisted mental health program for university students in China (CAMPUS): study protocol for a multi-center prospective cohort with planned embedded randomized intervention trials. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1877887
Who paid: The work was supported by Chinese funders including the National Natural Science Foundation of China, several Henan provincial programs and a Henan science and technology joint fund; the article states the funders had no role in study design, data collection, analysis, interpretation or manuscript preparation, and no equipment or software providers are credited.
Do not take this as professional medical advice.

