Doctors Are Using AI Nobody Approved — At Least in Sweden

survey · Journal of medical Internet research · la publicación, 28 jul 2026 · gratis

Médicos suecos usan ChatGPT sin autorización para diagnósticos y cartas a pacientes

Un estudio describe para qué lo usan, pero no puede decir cuántos lo hacen ni qué permiten las leyes de otros países.

Versión breve · la versión detallada sigue, unos 8 min

Pregunte a weeklyAI

Pregúnteme por este estudio: a quiénes se estudió, qué encontró y qué no dice.

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: Médicos que trabajan en servicios de salud
Cuántos: 357 médicos (de 557 invitados)
Dónde: Suecia
Cuándo: Entre diciembre de 2023 y enero de 2024
Tipo de estudio: survey
Quién lo hizo: Universidad de Halmstad, Suecia
El límite que importa: Es una foto de un solo momento y solo de Suecia; no dice qué pasa en su país.

Dónde trabajaban los médicos que respondieron

Hospitales públicos54%

Centros de salud públicos20%

Centros de salud privados14%

Porcentaje de los 357 médicos encuestados según su lugar de trabajo; no suman 100 porque hay otras categorías más pequeñas.

Esto es una fotografía de Suecia en ese período, no una descripción de lo que ocurre en su país.
Lectura de weeklyAI

Así podría verse · ilustración generada por weeklyAI.watch, no es una fotografía

Los médicos respondieron preguntas abiertas sobre cómo la inteligencia artificial había cambiado sus tareas. Al leer las respuestas, los investigadores se toparon con algo que no esperaban: varios contaban que usaban herramientas que nadie había aprobado en sus hospitales. Reiniciaron el análisis para estudiar ese patrón.

Así llegaron a lo que llaman "Shadow AI": el uso de herramientas de inteligencia artificial no autorizadas por médicos. En total, 357 médicos que trabajan en servicios de salud suecos respondieron la encuesta entre diciembre de 2023 y enero de 2024. La mayoría usaba aplicaciones de uso general, sobre todo ChatGPT, desde cuentas personales y dispositivos privados.

El estudio, publicado en el Journal of Medical Internet Research, describe cuatro propósitos. En el trabajo clínico, los médicos consultaban estas herramientas como un "colega" o una segunda opinión: para diagnósticos diferenciales, casos raros o síntomas difíciles de interpretar. En tareas administrativas, las usaban para traducir informes médicos complicados a un lenguaje que los pacientes entendieran. También las usaban para mantenerse al día con la investigación en su especialidad, y por simple curiosidad tecnológica, incluso programando sus propias herramientas.

Esto es una fotografía de Suecia en ese período, no una descripción de lo que ocurre en su país. El estudio no dice nada sobre lo que la ley de su país permite o exige hoy, ni sobre lo que su hospital autoriza.

Tampoco puede decirle qué tan común es la práctica. El estudio no fue diseñado para contar. Analizó respuestas escritas y las organizó en categorías; no ofrece porcentajes ni cifras sobre cuántos médicos hacen cada cosa. Además, quienes respondieron se ofrecieron voluntariamente y tenían interés en la tecnología, así que el panorama puede exagerar cuánto ocurre en realidad.

Los médicos también expresaron preocupación por el riesgo de depender demasiado de la herramienta y olvidar su propio conocimiento clínico, y porque los colegas más jóvenes no desarrollen su propia experiencia. El artículo también menciona riesgos para la privacidad de los datos, la seguridad clínica y el cumplimiento de normas. No concluye que la práctica sea segura ni beneficiosa en conjunto.

En Europa existe una razón regulatoria concreta: el software usado para diagnosticar o tratar se considera dispositivo médico y debe evaluarse antes de usarse. ChatGPT no pasó esa evaluación. Por eso los autores lo llaman uso no autorizado. Los investigadores sostienen que prohibirlo sin más probablemente no lo elimine.

Cuando un médico o un hospital usa una herramienta de inteligencia artificial con sus datos, usted puede preguntar qué herramienta es, quién la aprobó y qué pasa con su información. Puede empezar por su hospital o por la autoridad de protección de datos de su país.

¿Qué herramienta usa y quién la autorizó?

Qué significa para usted

Tenga presente que este estudio solo describe lo que relataron médicos en Suecia y no dice qué permiten las normas de su país ni de su institución. Si le preocupa cómo se usan sus datos en una consulta, pregunte qué herramienta se emplea, quién la autorizó y cómo se protege su información, y consulte a la autoridad de protección de datos correspondiente.

Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484

Quién pagó: La investigación fue financiada por la Fundación Sueca del Conocimiento a través del proyecto Business Models for Information-Driven Healthcare Ecosystems (subvención 220021) y la Multidisciplinary National Health Innovation Research School (subvención 20210047-H-02); los financiadores no tuvieron ningún papel en el diseño del estudio, la recogida de datos, el análisis, la interpretación ni la redacción.

No tome esto como consejo médico profesional.

Versión detalladaLos pasajes copiados del artículo, las ilustraciones y cada fuente con cuánto leímos de ella · 1528 palabras · unos 8 minLeerla →Cerrar

Médicos suecos usaron ChatGPT sin autorización para diagnosticar: qué significa para usted

Una encuesta a 357 médicos muestra que la IA ya entró a la consulta antes que la ley y el hospital la aprueben.

Un estudio publicado el 28 de julio de 2026 en el Journal of Medical Internet Research encontró que médicos suecos usan herramientas de inteligencia artificial no autorizadas en su trabajo. Los investigadores las agruparon en cuatro propósitos: trabajo clínico y toma de decisiones, trabajo administrativo, investigación y desarrollo profesional, e interés y curiosidad tecnológica¹. La encuesta se hizo entre diciembre de 2023 y enero de 2024 a 357 médicos que trabajaban en el sistema de salud de Suecia, de 557 invitados². La mayoría trabajaba en hospitales públicos, que fueron el grupo más grande con 194 de 357, seguidos por centros de salud públicos con 70 de 357 y centros de salud privados con 50 de 357³.

Para entender qué se está usando, conviene saber qué tecnología es. Los médicos describieron sobre todo aplicaciones de IA generativa de uso general, principalmente ChatGPT, a las que accedían con cuentas personales y dispositivos privados⁴. La IA generativa produce texto con apariencia humana: usted le escribe una pregunta y le devuelve una respuesta redactada. El problema regulatorio es concreto. Según el Reglamento Europeo de Productos Sanitarios, las herramientas de IA usadas con fines clínicos deben pasar una evaluación de conformidad antes de usarse; las herramientas de propósito general como ChatGPT no lo han hecho, lo que vuelve su uso clínico no autorizado a nivel regulatorio⁵. Esas herramientas no llevan el marcado europeo de conformidad, no pasaron la evaluación y no están clasificadas como productos sanitarios⁶.

¿Cómo se veía ese uso en la práctica? Un médico relató que ingresó datos clínicos sin identificar al paciente —historia médica, hallazgos de examen, resultados de pruebas— en ChatGPT, que le sugirió varios diagnósticos diferenciales⁷. Otro dijo que usó ChatGPT en ocasiones para casos raros y posibles diagnósticos⁸. Un tercero contó que para generar cartas a pacientes ingresó información de informes radiológicos complejos en GPT, que la resumió de forma más fácil de entender⁹. Y otro más relató que programó sus propias herramientas, sobre todo con ChatGPT, para obtener información rápido antes de decidir¹⁰.

Los propios autores marcan los límites. La recolección de datos fue transversal: una foto de un solo momento, y existe un sesgo de autoselección, porque los médicos interesados en tecnología son más propensos a responder¹¹. El estudio se hizo solo en Suecia, y ese foco en un único país puede limitar que los hallazgos se trasladen a otros países¹². Es un análisis cualitativo de respuestas abiertas, no un conteo: no puede decirle qué tan común es la práctica donde usted vive.

El problema no es exclusivo de Suecia. Según el resumen de una encuesta a países de la región europea de la Organización Mundial de la Salud —pudimos leer solo el resumen, el texto completo está detrás de una suscripción—, de 50 Estados miembros que respondieron, 4 de 50 tienen una estrategia de IA específica para salud y 7 están desarrollando una¹³. Casi la mitad, 23, reportó evaluaciones en curso de leyes y políticas sobre sistemas de IA, y 10 desarrollaron nuevas leyes de IA específicas para salud¹⁴. Solo 14 emitieron guías sobre las implicaciones éticas del uso de IA en salud¹⁵. Menos del 10%, 4 Estados, desarrollaron estándares de responsabilidad para IA¹⁶. La conclusión de ese trabajo es que la gobernanza de la IA en salud está poco desarrollada en la región¹⁷.

El mismo resumen de otro trabajo sobre cirugía —también leído solo en su resumen— señala que la IA se usa cada vez más en atención quirúrgica para apoyo a decisiones, planificación de operaciones, guía intraoperatoria y funciones autónomas¹⁸. Esos sistemas pueden mejorar la eficiencia y el desempeño clínico, pero introducen riesgos tecnológicos, humanos, legales y éticos¹⁹. Y los marcos regulatorios y legales actuales no están del todo equipados para responder a la IA quirúrgica²⁰. Los riesgos pueden manifestarse como error diagnóstico, error de tratamiento, consentimiento informado comprometido, erosión de la confianza del paciente y violaciones de privacidad²¹.

Hay un tercer trabajo que apunta a lo mismo desde otro ángulo. Según su resumen —leído solo en su resumen—, un experimento con 768 estudiantes universitarios probó dos mensajes: uno de garantía de privacidad y otro de advertencia sobre los límites profesionales²². El mensaje de privacidad aumentó la percepción de protección²³, y la advertencia sobre límites profesionales aumentó la conciencia de esos límites²⁴. La confianza calibrada fue más alta cuando ambos mensajes estaban presentes²⁵. Los autores concluyen que esos mensajes no deben entenderse como una invitación a usar más los chatbots, sino como ayuda para tratar estas herramientas como lo que son: instrumentos limitados²⁶.

Así lo leemos nosotros. Cuando no está claro quién manda, es fácil confundir el ejercicio razonable de un criterio profesional con un abuso, y la libertad con hacer lo que sea. En un caso así esperaríamos que los médicos no vivan su uso de ChatGPT como una falta grave, sino como parte de su buen juicio: actúo por el bien del paciente y las reglas formales van lentas. Sabríamos que nos equivocamos si los propios médicos dijeran que están desobedeciendo una norma clara y sintieran culpa o miedo, o si dejaran de usar la herramienta por respeto a la autoridad que la prohíbe. Usted puede hacer algo con esto: cuando escuche a un profesional decir que usa una herramienta no aprobada "por el bien del paciente", pregúntese si esa misma justificación serviría para saltarse otras reglas. Y exija que las autoridades expliquen con claridad qué está permitido y qué no. La confusión entre autoridad legítima y abuso se combate con reglas claras, no con silencio.

Hay una segunda cosa que nos parece importante. Quien ejerce un poder delegado debería rendir cuentas ante quienes representa. En este caso, los médicos que usan IA no autorizada ocupan un espacio de decisión clínica sin control visible de la institución ni del paciente. Esperaríamos que no existan canales claros para reclamar por una decisión médica apoyada en una IA no aprobada, y que la responsabilidad quede difusa entre el médico, el hospital y el fabricante de la herramienta. Sabríamos que nos equivocamos si esos canales existieran y fueran conocidos, o si los médicos rindieran cuentas explícitas ante un comité de ética por ese uso. Usted puede preguntar en su hospital o clínica quién responde si una decisión médica se apoya en una IA no autorizada. Si nadie sabe, eso ya le dice algo. Y puede pedir que se aclaren los canales de reclamación.

Una tercera observación, y quizá la más práctica. Las personas se acostumbran a obedecer y dejan de pensar por sí mismas; con el tiempo, esa costumbre puede hacer que dependan de una autoridad externa, aunque esa autoridad sea una máquina. Los propios médicos del estudio advirtieron que los colegas jóvenes podrían no desarrollar experiencia clínica propia si se acostumbran a consultar la IA. Esperaríamos que, con los años, quienes usan estas herramientas pierdan confianza en su propio juicio diagnóstico y consulten la máquina incluso en casos sencillos, sin verificar. Sabríamos que nos equivocamos si los médicos las usaran solo como apoyo ocasional, mantuvieran su criterio intacto y verificaran siempre con otras fuentes. Usted puede hacer algo concreto: si un médico le dice que consultó una IA para su diagnóstico, pregúntele cómo verificó esa información con su propio criterio o con otras fuentes. Y si usted usa IA en su trabajo, revise si está pensando por sí mismo o solo aceptando lo que la herramienta le dice.

¿Qué haría falta para que esto cambie? Los autores del estudio sueco plantean que las organizaciones de salud podrían beneficiarse de reconocer la realidad de la IA en la sombra y crear entornos controlados y transparentes para experimentar: espacios de prueba clínicos, rutas de testeo supervisadas o estructuras de apoyo dedicadas a la IA²⁷. Y sostienen que los líderes de salud no deberían ver estas prácticas solo como una violación de cumplimiento, sino como una fuente de innovación que señala necesidades profesionales no atendidas²⁸.

Otro trabajo sobre gobernanza —leído solo en su resumen— pide mecanismos legales y de política adaptativos para responder a los desafíos cambiantes de integrar IA en los sistemas de salud²⁹. Y el trabajo sobre cirugía concluye que la integración segura de IA requiere más que desempeño técnico: hacen falta gobernanza sólida, vigilancia continua del desempeño, rutas de respuesta ante incidentes, credenciales para los clínicos y participación de las sociedades de especialidad³⁰. Ahí está la tensión: lo mismo que hace valiosa la herramienta para el médico es lo que la vuelve riesgosa para el sistema.

Para terminar, esto es lo que queda en sus manos. La próxima vez que vaya a una consulta —en su país, con su sistema de salud, con sus reglas— puede preguntar si el profesional usó alguna herramienta de IA para orientar su caso, y si esa herramienta está aprobada por la autoridad sanitaria local. También puede preguntar quién es responsable si esa orientación resulta equivocada. No es una pregunta técnica: es la pregunta de quién responde. Y si la respuesta es "nadie sabe", eso ya es información que usted necesitaba tener.

De dónde sale cada dato de contexto, y cuánto leímos de cada documento

Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Physicians use Shadow AI for several purposes, which we grouped into 4 categories: clinical work and decision-making, administrative work, research and professional development, and technological interest and curiosity."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - el artículo del que trata esta nota — el artículo completo — el pasaje: "A total of 557 physicians meeting the eligibility criteria were invited to participate, yielding 357 complete responses (response rate~64%)."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Most respondents worked in public sector settings, with public hospitals representing the largest group (194/357, 54%), followed by public health centers (70/357, 20%) and private health centers (50/357, 14%)."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - el artículo del que trata esta nota — el artículo completo — el pasaje: "the physicians’ descriptions of their use of AI were predominantly general-purpose generative AI applications, mainly ChatGPT, accessed through personal accounts and private devices."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Under the European Union Medical Device Regulation, AI tools used for clinical purposes must undergo conformity assessment before use; general-purpose tools such as ChatGPT have not done so, rendering their clinical application unauthorized at the regulatory level."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - el artículo del que trata esta nota — el artículo completo — el pasaje: "General-purpose generative AI tools such as ChatGPT are not Conformité Européenne marked, have not undergone conformity assessment, and are not classified as medical devices."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - el artículo del que trata esta nota — el artículo completo — el pasaje: "I entered de-identified data (medical history, examination [or observed findings/clinical findings], test results) into ChatGPT, which suggested various differential diagnoses” [Participant 58]."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - el artículo del que trata esta nota — el artículo completo — el pasaje: "I have used ChatGPT on occasion to get help with rare conditions in order to suggest possible diagnoses” [Participant 202]."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - el artículo del que trata esta nota — el artículo completo — el pasaje: "To generate letters for patients, information from complex radiology reports was entered into GPT, which summarized the information in patient letters in a way that is easier for them to understand ."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - el artículo del que trata esta nota — el artículo completo — el pasaje: "I have programmed tools myself, mostly via ChatGPT, but also other open-source AI tools. I use these tools to quickly obtain information before decision-making, when appropriate ."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - el artículo del que trata esta nota — el artículo completo — el pasaje: "the cross-sectional nature of our data collection offers a snapshot of AI adoption experiences at a single point in time. Additionally, there is a potential self-selection bias; physicians interested in technology are more likely to respond."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - el artículo del que trata esta nota — el artículo completo — el pasaje: "although our sample included physicians from diverse health care settings across Sweden, the study’s exclusive focus on a single national context may limit the transferability of our findings to other countries."
Adib K, Letchford N, Dunning HE, Salama N, Tolias Y, De Barros J, et al. (2026). Governance of artificial intelligence for health systems, WHO European Region. Bulletin of the World Health Organization. 10.2471/blt.25.294978 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Of the 50 Member States responding to the survey, 8% (4/50) have a health-specific AI strategy and 14% (7) are developing one."
Adib K, Letchford N, Dunning HE, Salama N, Tolias Y, De Barros J, et al. (2026). Governance of artificial intelligence for health systems, WHO European Region. Bulletin of the World Health Organization. 10.2471/blt.25.294978 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Nearly half the Member States (23) reported ongoing assessments of laws and policies on AI systems and a fifth (10) have developed new health-specific AI laws."
Adib K, Letchford N, Dunning HE, Salama N, Tolias Y, De Barros J, et al. (2026). Governance of artificial intelligence for health systems, WHO European Region. Bulletin of the World Health Organization. 10.2471/blt.25.294978 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Only 14 Member States have issued guidelines to address the ethical implications of using AI in health or across sectors."
Adib K, Letchford N, Dunning HE, Salama N, Tolias Y, De Barros J, et al. (2026). Governance of artificial intelligence for health systems, WHO European Region. Bulletin of the World Health Organization. 10.2471/blt.25.294978 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Less than 10% (4) of Member States have developed liability standards for AI or guidance on the application of existing liability standards."
Adib K, Letchford N, Dunning HE, Salama N, Tolias Y, De Barros J, et al. (2026). Governance of artificial intelligence for health systems, WHO European Region. Bulletin of the World Health Organization. 10.2471/blt.25.294978 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "In the WHO European Region, governance of AI in health care is underdeveloped."
Hashimoto DA, Marwaha JS, Lee SA, Schwaitzberg S, Duffourc MN. (2026). Risk and liability in the deployment of AI systems for surgery: a SAGES white paper. Surgical Endoscopy. 10.1007/s00464-026-12881-8 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Artificial intelligence (AI) is increasingly utilized in surgical care for decision support, operative planning, intraoperative guidance, and autonomous functions."
Hashimoto DA, Marwaha JS, Lee SA, Schwaitzberg S, Duffourc MN. (2026). Risk and liability in the deployment of AI systems for surgery: a SAGES white paper. Surgical Endoscopy. 10.1007/s00464-026-12881-8 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "While these systems can enhance efficiency and clinical performance, they also introduce risks related to technology, human factors, legal issues, and ethics."
Hashimoto DA, Marwaha JS, Lee SA, Schwaitzberg S, Duffourc MN. (2026). Risk and liability in the deployment of AI systems for surgery: a SAGES white paper. Surgical Endoscopy. 10.1007/s00464-026-12881-8 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Current regulatory and legal frameworks are not fully equipped to address the challenges of AI-assisted surgery."
Hashimoto DA, Marwaha JS, Lee SA, Schwaitzberg S, Duffourc MN. (2026). Risk and liability in the deployment of AI systems for surgery: a SAGES white paper. Surgical Endoscopy. 10.1007/s00464-026-12881-8 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "These risks may manifest clinically as diagnostic error, treatment error, compromised informed consent, erosion of patient trust, threats to therapeutic autonomy, and privacy violations."
Zhang Z, Lu X, Zhang Y, Zhang H, Zhang M. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. 10.3389/fpsyg.2026.1934264 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "We conducted a 2 × 2 randomized vignette experiment with 768 college students."
Zhang Z, Lu X, Zhang Y, Zhang H, Zhang M. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. 10.3389/fpsyg.2026.1934264 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Privacy assurance increased perceived privacy protection, F (1, 764) = 159.30, p 2 = 0.172, d = 0.91."
Zhang Z, Lu X, Zhang Y, Zhang H, Zhang M. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. 10.3389/fpsyg.2026.1934264 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Professional-boundary warning increased boundary awareness, F (1, 764) = 176.40, p 2 = 0.188, d = 0.96."
Zhang Z, Lu X, Zhang Y, Zhang H, Zhang M. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. 10.3389/fpsyg.2026.1934264 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Calibrated trust was highest when both messages were present."
Zhang Z, Lu X, Zhang Y, Zhang H, Zhang M. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. 10.3389/fpsyg.2026.1934264 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Such messages should not be understood as prompts for greater use. Rather, they may help users treat chatbots as limited tools for information and support navigation and recognize when professional help is needed."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - el artículo del que trata esta nota — el artículo completo — el pasaje: "health care organizations may benefit from acknowledging the reality of Shadow AI and developing controlled and transparent environments for experimentation, such as clinical sandboxes, supervised testing pathways, or dedicated AI support structures."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - el artículo del que trata esta nota — el artículo completo — el pasaje: "We argue that health care leaders should not view Shadow AI solely as a compliance violation but as a source of user-driven innovation that signals unmet professional needs."
Adib K, Letchford N, Dunning HE, Salama N, Tolias Y, De Barros J, et al. (2026). Governance of artificial intelligence for health systems, WHO European Region. Bulletin of the World Health Organization. 10.2471/blt.25.294978 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Adaptive legal and policy mechanisms are needed to respond effectively to the complex and evolving challenges of AI integration in health systems."
Hashimoto DA, Marwaha JS, Lee SA, Schwaitzberg S, Duffourc MN. (2026). Risk and liability in the deployment of AI systems for surgery: a SAGES white paper. Surgical Endoscopy. 10.1007/s00464-026-12881-8 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Robust governance, ongoing performance surveillance, incident response pathways, clinician credentialing, and specialty-society engagement are needed to reduce harm and clarify accountability."

No tome esto como consejo médico profesional.

survey · Journal of medical Internet research · the paper, 28 Jul 2026 · free

Doctors Are Using AI Nobody Approved — At Least in Sweden

Survey of 357 Swedish physicians found unauthorized ChatGPT use for diagnoses, patient letters and research. It cannot say how common this is anywhere else, or what your own country allows.

Short version · the longer version follows, about 7 min

Ask weeklyAI

Ask me about this study: who was studied, what it found, and what it does not say.

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: Physicians working in Swedish health care
How many: 357 complete responses
Where: Sweden
When: December 2023 to January 2024
Kind of study: survey
Who did it: Halmstad University, Sweden
The limit that matters: Only Sweden, only two months, and those interested in technology were more likely to answer.

Where the 357 physicians worked

Public hospitals54%

Public health centers20%

Private health centers14%

Share of the 357 responding physicians in each setting; the remaining physicians worked in other settings not shown here.

The next time you are at a hospital or clinic, you can ask the staff: what rules cover AI tools used on my medical information, and who approved them?
weeklyAI's reading

How it could look · illustration generated by weeklyAI.watch, not a photograph

In Sweden, physicians answered a questionnaire about how they use artificial intelligence at work. A surprising thing surfaced in their written answers: some described reaching for tools their hospitals had never approved.

The small study, published in the Journal of Medical Internet Research, was conducted by Lena Petersson, Luís Irgang, Ingela Mauritzon, and Magnus Holmén from Halmstad University, and drew 357 physicians employed in Swedish health care organizations. They responded between December 2023 and January 2024. All had earlier confirmed they used AI in their work.

What they described was unauthorized. Under European Union rules, software used for diagnosis, prediction, monitoring or treatment counts as a medical device and needs a conformity assessment before clinical use. General-purpose tools like ChatGPT have not undergone that assessment. The physicians in this study mostly used such tools through personal accounts and private devices. The researchers call this Shadow AI.

In their free-text answers, physicians said they used these tools as a colleague or second opinion — to discuss possible diagnoses, to look up rare conditions, to work through hard-to-interpret symptoms. They used them to turn complex medical language into letters patients could understand. They used them to keep up with research. Some did it out of plain curiosity about what the technology could do.

The study is only a snapshot of Sweden during those two months. The study's exclusive focus on a single national context may limit the transferability of its findings to other countries. Further research is needed to determine whether the results from this study are transferable to other countries.

It also cannot tell you how common this practice is — anywhere. The researchers did not set out to count. They read written answers and grouped them into four categories: clinical work and decision-making, administrative work, research and professional development, and technological interest and curiosity. No percentage is attached to any of them.

And the picture may be skewed. Physicians chose whether to take part, and those interested in technology were more likely to respond. The article notes potential self-selection bias because physicians interested in technology were more likely to respond.

The researchers also report what physicians themselves worried about: that they might lean too heavily on AI and forget their own clinical knowledge, that younger colleagues might never build experience of their own, that there is little time to learn these tools and little technical support nearby.

What does this offer you where you live? A question you can actually ask. The next time you are at a hospital or clinic, you can ask the staff: what rules cover AI tools used on my medical information, and who approved them? That question is yours to ask, and it is free.

What this means for you

What happened in Sweden does not tell you what your own country permits, so treat it as one question worth carrying, not a conclusion. When you next deal with a clinic or hospital, you can ask who approved the AI tools touching your medical information and under what rules.

Who paid: The research was funded by the Swedish Knowledge Foundation through the Business Models for Information-Driven Healthcare Ecosystems project (grant 220021) and the Multidisciplinary National Health Innovation Research School (grant 20210047-H-02); the funders had no role in study design, data collection, analysis, interpretation, or writing.

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 · 1323 words · about 7 minRead it →Close

Swedish Doctors Used Unapproved AI on Patient Data, Study Finds

The survey asked physicians what they used unauthorized tools for. It did not measure whether patients were helped or harmed.

In a survey of 357 physicians working in Swedish health care, researchers found that doctors described using unauthorized artificial intelligence tools—mostly ChatGPT—for clinical decisions, administrative work, research and professional development, and out of curiosity. The data were collected between December 2023 and January 2024.

The survey invited 557 physicians and received 357 complete responses, a response rate of about 64 percent¹. Most worked in public hospitals (194 of 357, or 54 percent), followed by public health centers (70, or 20 percent) and private health centers (50, or 14 percent)². The researchers grouped what physicians described into four categories: clinical work and decision-making, administrative work, research and professional development, and technological interest and curiosity³.

The physicians said they used these tools as a "colleague" or "second opinion" for diagnosis. One wrote: "I entered de-identified data (medical history, examination [or observed findings/clinical findings], test results) into ChatGPT, which suggested various differential diagnoses"⁴. Another used it "to get help with rare conditions in order to suggest possible diagnoses"⁵. For administrative work, one physician described entering complex radiology reports into GPT so it could summarize the information in patient letters "in a way that is easier for them to understand"⁶. Some went further: "I have programmed tools myself, mostly via ChatGPT, but also other open-source AI tools"⁷.

What makes this "Shadow AI" is not that the tools are secret in any dramatic sense. It is that they are unauthorized. Under the European Union Medical Device Regulation, AI tools used for clinical purposes must undergo conformity assessment before use⁸. General-purpose tools like ChatGPT have not done so, are not Conformité Européenne marked, and are not classified as medical devices⁹. When a physician uses ChatGPT to think through a diagnosis, the tool has no legal basis for that use.

The physicians in this study accessed these tools "predominantly" through personal accounts and private devices¹⁰. They were not procured through hospital channels, not integrated into clinical systems, and not subject to organizational oversight. The classification as Shadow AI rests on two conditions being absent at once: regulatory approval and institutional endorsement.

This is not the first time unauthorized technology has entered medical workplaces. The pattern resembles what researchers call Shadow IT—messaging apps, cloud storage, informal communication systems that employees adopt when official tools feel slow or ill-suited. The study's authors compare the two, noting that Shadow AI introduces something new: generative systems capable of producing reasoning, predictions and clinical suggestions, not just storing or transmitting information.

The regulatory gap is not unique to Sweden. A survey of 50 member states in the WHO European Region found that only 8 percent (4 of 50) have a health-specific AI strategy, and 14 percent (7) are developing one¹¹. Nearly half (23) reported ongoing assessments of laws and policies on AI systems, and a fifth (10) have developed new health-specific AI laws¹². Only 14 member states have issued guidelines addressing the ethical implications of using AI in health or across sectors¹³. Less than 10 percent (4) have developed liability standards for AI or guidance on applying existing standards¹⁴. The summary of that study—we could read only the summary, the full paper is behind a subscription—concluded that governance of AI in health care in the region is "underdeveloped"¹⁵.

The risks of AI in clinical settings are not hypothetical. A white paper from a surgical society—we could read only the summary—describes how AI is already used in surgery for decision support, operative planning, intraoperative guidance and autonomous functions¹⁶. It warns that current regulatory and legal frameworks are "not fully equipped" to address the challenges¹⁷. The paper proposes that risks come from three sources: the AI system itself, the clinician using it, and the institution deploying it¹⁸. These risks may show up as diagnostic error, treatment error, compromised informed consent, erosion of patient trust and privacy violations¹⁹. Liability, the authors note, may extend not only to surgeons but also to developers for defective design and to institutions for negligent implementation or oversight²⁰.

The Swedish study's own authors acknowledge limits. The cross-sectional design offers only a snapshot at a single point in time, and physicians interested in technology were more likely to respond²¹. The exclusive focus on Sweden may limit how well the findings transfer to other countries²². The free-text format captured purposes and motivations but rarely included detail about how physicians actually interacted with the tools—what they typed, how they checked answers, whether they verified outputs against other sources.

The study did not measure whether patients were helped or harmed. It did not count how often Shadow AI was used. It did not test whether prohibiting these tools would stop their use. What it found is what physicians said they did.

Here is how we read it. The physicians in this study were not rogue actors. They were professionals trying to do their jobs in systems that had not yet given them approved tools for the tasks they faced. When a doctor asks an AI for help thinking through a rare diagnosis, the question is not just whether the tool is legal. It is whether the doctor treats the answer as a suggestion or as an authority. The study cannot tell us which. And that distinction matters more than the regulation itself.

We also read the pattern differently than a compliance officer might. The same conditions that make Shadow AI risky—no oversight, no verification protocols, no documentation—are the conditions that make it professionally valuable. A tool you can reach instantly, without waiting for procurement, fills a gap. If hospitals want physicians to stop using unauthorized AI, they will need to offer something that works as fast. The study's authors suggest clinical sandboxes and supervised testing pathways²³. Whether those would satisfy a physician mid-diagnosis is an open question.

We also read a study of college students that tested how people respond to privacy assurances and warnings about professional boundaries—we could read only the summary. It found that trust was highest when both messages were present, and that boundary awareness was linked to lower overreliance and stronger willingness to seek professional help²⁴²⁵. The authors cautioned that such messages should not be understood as prompts for greater use, but as ways to help people treat chatbots as limited tools and recognize when professional help is needed²⁶. We cannot say whether those findings apply to physicians, but they suggest that how a tool is framed matters.

Other work we read—again, only the summary—argues that safe integration of AI into surgery requires more than technical performance. It calls for robust governance, ongoing performance surveillance, incident response pathways, clinician credentialing and engagement from professional societies²⁷. The summary of a governance survey in the WHO European Region concluded that adaptive legal and policy mechanisms are needed to respond to the evolving challenges of AI in health systems²⁸. These are prescriptions, not findings. They describe what experts think should happen, not what has been shown to work.

For readers in Latin America, the United States or Canada, this study says nothing about what your own country's law permits. It is a snapshot of Swedish physicians in a Swedish system. But it raises a question you can ask anywhere: when your doctor uses a tool to help think through your case, what is that tool, who approved it, and what happens to the information you shared? You do not need to know the answer to ask.

The most useful thing you can take from this is not a verdict on AI in medicine. It is a question to bring to your own care. The next time you sit in an examination room, you can ask whether any AI tool was used in your case and how the results were checked. That question is yours to ask, regardless of what any regulator has or has not decided.

Where each piece of context comes from, and how much of it we read

Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - the article this story is about — the whole article — the passage: "A total of 557 physicians meeting the eligibility criteria were invited to participate, yielding 357 complete responses (response rate~64%)."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - the article this story is about — the whole article — the passage: "Most respondents worked in public sector settings, with public hospitals representing the largest group (194/357, 54%), followed by public health centers (70/357, 20%) and private health centers (50/357, 14%)."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - the article this story is about — the whole article — the passage: "Physicians use Shadow AI for several purposes, which we grouped into 4 categories: clinical work and decision-making, administrative work, research and professional development, and technological interest and curiosity."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - the article this story is about — the whole article — the passage: "I entered de-identified data (medical history, examination [or observed findings/clinical findings], test results) into ChatGPT, which suggested various differential diagnoses” [Participant 58]."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - the article this story is about — the whole article — the passage: "I have used ChatGPT on occasion to get help with rare conditions in order to suggest possible diagnoses” [Participant 202]."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - the article this story is about — the whole article — the passage: "To generate letters for patients, information from complex radiology reports was entered into GPT, which summarized the information in patient letters in a way that is easier for them to understand ."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - the article this story is about — the whole article — the passage: "I have programmed tools myself, mostly via ChatGPT, but also other open-source AI tools. I use these tools to quickly obtain information before decision-making, when appropriate ."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - the article this story is about — the whole article — the passage: "Under the European Union Medical Device Regulation, AI tools used for clinical purposes must undergo conformity assessment before use; general-purpose tools such as ChatGPT have not done so, rendering their clinical application unauthorized at the regulatory level."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - the article this story is about — the whole article — the passage: "General-purpose generative AI tools such as ChatGPT are not Conformité Européenne marked, have not undergone conformity assessment, and are not classified as medical devices."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - the article this story is about — the whole article — the passage: "the physicians’ descriptions of their use of AI were predominantly general-purpose generative AI applications, mainly ChatGPT, accessed through personal accounts and private devices."
Adib K, Letchford N, Dunning HE, Salama N, Tolias Y, De Barros J, et al. (2026). Governance of artificial intelligence for health systems, WHO European Region. Bulletin of the World Health Organization. 10.2471/blt.25.294978 — only the abstract - the full text could not be fetched — the passage: "Of the 50 Member States responding to the survey, 8% (4/50) have a health-specific AI strategy and 14% (7) are developing one."
Adib K, Letchford N, Dunning HE, Salama N, Tolias Y, De Barros J, et al. (2026). Governance of artificial intelligence for health systems, WHO European Region. Bulletin of the World Health Organization. 10.2471/blt.25.294978 — only the abstract - the full text could not be fetched — the passage: "Nearly half the Member States (23) reported ongoing assessments of laws and policies on AI systems and a fifth (10) have developed new health-specific AI laws."
Adib K, Letchford N, Dunning HE, Salama N, Tolias Y, De Barros J, et al. (2026). Governance of artificial intelligence for health systems, WHO European Region. Bulletin of the World Health Organization. 10.2471/blt.25.294978 — only the abstract - the full text could not be fetched — the passage: "Only 14 Member States have issued guidelines to address the ethical implications of using AI in health or across sectors."
Adib K, Letchford N, Dunning HE, Salama N, Tolias Y, De Barros J, et al. (2026). Governance of artificial intelligence for health systems, WHO European Region. Bulletin of the World Health Organization. 10.2471/blt.25.294978 — only the abstract - the full text could not be fetched — the passage: "Less than 10% (4) of Member States have developed liability standards for AI or guidance on the application of existing liability standards."
Adib K, Letchford N, Dunning HE, Salama N, Tolias Y, De Barros J, et al. (2026). Governance of artificial intelligence for health systems, WHO European Region. Bulletin of the World Health Organization. 10.2471/blt.25.294978 — only the abstract - the full text could not be fetched — the passage: "In the WHO European Region, governance of AI in health care is underdeveloped."
Hashimoto DA, Marwaha JS, Lee SA, Schwaitzberg S, Duffourc MN. (2026). Risk and liability in the deployment of AI systems for surgery: a SAGES white paper. Surgical Endoscopy. 10.1007/s00464-026-12881-8 — only the abstract - the full text could not be fetched — the passage: "Artificial intelligence (AI) is increasingly utilized in surgical care for decision support, operative planning, intraoperative guidance, and autonomous functions."
Hashimoto DA, Marwaha JS, Lee SA, Schwaitzberg S, Duffourc MN. (2026). Risk and liability in the deployment of AI systems for surgery: a SAGES white paper. Surgical Endoscopy. 10.1007/s00464-026-12881-8 — only the abstract - the full text could not be fetched — the passage: "Current regulatory and legal frameworks are not fully equipped to address the challenges of AI-assisted surgery."
Hashimoto DA, Marwaha JS, Lee SA, Schwaitzberg S, Duffourc MN. (2026). Risk and liability in the deployment of AI systems for surgery: a SAGES white paper. Surgical Endoscopy. 10.1007/s00464-026-12881-8 — only the abstract - the full text could not be fetched — the passage: "Risks associated with surgical AI can be understood through a tripartite framework: risks inherent to the AI system, risks introduced by the clinician-user, and risks arising from institutional deployment."
Hashimoto DA, Marwaha JS, Lee SA, Schwaitzberg S, Duffourc MN. (2026). Risk and liability in the deployment of AI systems for surgery: a SAGES white paper. Surgical Endoscopy. 10.1007/s00464-026-12881-8 — only the abstract - the full text could not be fetched — the passage: "These risks may manifest clinically as diagnostic error, treatment error, compromised informed consent, erosion of patient trust, threats to therapeutic autonomy, and privacy violations."
Hashimoto DA, Marwaha JS, Lee SA, Schwaitzberg S, Duffourc MN. (2026). Risk and liability in the deployment of AI systems for surgery: a SAGES white paper. Surgical Endoscopy. 10.1007/s00464-026-12881-8 — only the abstract - the full text could not be fetched — the passage: "Although surgeons remain the ultimate clinical decision-makers, liability may also extend to developers for defective design or failure to warn, and to institutions for negligent implementation or oversight."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - the article this story is about — the whole article — the passage: "the cross-sectional nature of our data collection offers a snapshot of AI adoption experiences at a single point in time. Additionally, there is a potential self-selection bias; physicians interested in technology are more likely to respond."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - the article this story is about — the whole article — the passage: "although our sample included physicians from diverse health care settings across Sweden, the study’s exclusive focus on a single national context may limit the transferability of our findings to other countries."
Petersson, L., Irgang, L., Mauritzon, I. et al. (2026). Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians’ Free-Text Answers. Journal of Medical Internet Research. https://doi.org/10.2196/93484 - the article this story is about — the whole article — the passage: "health care organizations may benefit from acknowledging the reality of Shadow AI and developing controlled and transparent environments for experimentation, such as clinical sandboxes, supervised testing pathways, or dedicated AI support structures."
Zhang Z, Lu X, Zhang Y, Zhang H, Zhang M. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. 10.3389/fpsyg.2026.1934264 — only the abstract - the full text could not be fetched — the passage: "Calibrated trust was highest when both messages were present."
Zhang Z, Lu X, Zhang Y, Zhang H, Zhang M. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. 10.3389/fpsyg.2026.1934264 — only the abstract - the full text could not be fetched — the passage: "Calibrated trust and perceived privacy protection were associated with safe-use intention, whereas boundary awareness was linked to lower overreliance risk and stronger professional help-seeking intention."
Zhang Z, Lu X, Zhang Y, Zhang H, Zhang M. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. 10.3389/fpsyg.2026.1934264 — only the abstract - the full text could not be fetched — the passage: "Such messages should not be understood as prompts for greater use. Rather, they may help users treat chatbots as limited tools for information and support navigation and recognize when professional help is needed."
Hashimoto DA, Marwaha JS, Lee SA, Schwaitzberg S, Duffourc MN. (2026). Risk and liability in the deployment of AI systems for surgery: a SAGES white paper. Surgical Endoscopy. 10.1007/s00464-026-12881-8 — only the abstract - the full text could not be fetched — the passage: "Robust governance, ongoing performance surveillance, incident response pathways, clinician credentialing, and specialty-society engagement are needed to reduce harm and clarify accountability."
Adib K, Letchford N, Dunning HE, Salama N, Tolias Y, De Barros J, et al. (2026). Governance of artificial intelligence for health systems, WHO European Region. Bulletin of the World Health Organization. 10.2471/blt.25.294978 — only the abstract - the full text could not be fetched — the passage: "Adaptive legal and policy mechanisms are needed to respond effectively to the complex and evolving challenges of AI integration in health systems."

Do not take this as professional medical advice.