The Form You Sign for an AI Trial Often Doesn't Say What the AI Does

analysis of texts · Journal of medical Internet research · la publicación, 2 jul 2026 · gratis

Más de la mitad de los formularios de ensayos con IA no dicen qué IA usan

Un análisis de 114 ensayos clínicos con documentos de consentimiento públicos halló que muchos callan el tipo de inteligencia artificial, los riesgos y el destino de los datos; sólo cubre un registro internacional.

Versión breve · la versión detallada sigue, unos 7 min

Pregunte al chat de 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: Documentos de consentimiento informado de ensayos clínicos con inteligencia artificial
Cuántos: 114
Dónde: ClinicalTrials.gov
Cuándo: Búsqueda realizada el 21 de junio de 2025
Tipo de estudio: analysis of what people did
Quién lo hizo: Hospital Xiangya, Universidad Central del Sur, China
El límite que importa: Solo cubre documentos públicos de un registro; no midió si los participantes entendieron.

Qué dicen los formularios sobre los datos si usted se retira

No dicen nada44.7%

Dicen que los datos se destruyen26.3%

Dicen que los datos ya recogidos se siguen usando25.4%

Le ofrecen elegir3.5%

Porcentaje de los 114 formularios de consentimiento según qué dicen sobre el manejo de los datos después del retiro; los porcentajes no suman 100 porque algunos no encajan en estas categorías.

Cuando le entreguen un formulario para un ensayo con inteligencia artificial, busque tres cosas: qué tipo de IA se usa y para qué, qué riesgos existen, y qué pasará con sus datos si decide retirarse.
Lectura de weeklyAI

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

Investigadores del Hospital Xiangya, en China, revisaron 114 ensayos clínicos que usan inteligencia artificial y que tienen su documento de consentimiento publicado en ClinicalTrials.gov. La búsqueda se hizo el 21 de junio de 2025.

En cada formulario revisaron tres cosas: la información básica que da, qué tan fácil es de leer y qué pasa con los datos de quien participa.

El resultado es incómodo. Más de la mitad, 66 de los 114, nunca dice qué tipo de inteligencia artificial se usa ni para qué sirve. Y 21 formularios, el 18.4%, no mencionan ningún riesgo.

La lectura tampoco ayuda. Sólo 16 formularios, el 14%, cumplen las dos condiciones que los autores consideran básicas: ser cortos (menos de 15,000 caracteres) y tener una dificultad de lectura por debajo del nivel universitario. Los ensayos de mayor riesgo no tenían formularios más fáciles de entender.

Los datos personales quedan en un limbo. Ante la pregunta de qué pasa con la información si usted se retira, 51 formularios (44.7%) no dicen nada. Treinta indican que los datos se destruyen, 29 que se siguen usando, y sólo 4 (3.5%) le ofrecen elegir. Las leyes de protección citadas varían y ninguna predomina.

Los autores proponen una lista de requisitos mínimos para estos consentimientos: decir el tipo de IA y su nivel de autonomía, presentar los riesgos con la misma importancia que los beneficios, acortar y simplificar el texto, incluir al menos un gráfico que ayude a entender, y explicar con claridad el manejo de los datos después del retiro.

Esa lista es una propuesta. No fue probada en ningún ensayo, así que no se sabe si mejora los consentimientos.

Hay más límites. El estudio sólo miró ClinicalTrials.gov, el único registro que publica los documentos de consentimiento. No se sabe si los resultados serían iguales en otros registros. Tampoco analizó los documentos finales que se firman en el sitio, que pueden ser distintos, ni midió si las personas realmente entendieron lo que leyeron.

Cuando le entreguen un formulario para un ensayo con inteligencia artificial, busque tres cosas: qué tipo de IA se usa y para qué, qué riesgos existen, y qué pasará con sus datos si decide retirarse.

Qué significa para usted

A usted, que quizá firme un formulario de un ensayo con inteligencia artificial, le conviene preguntar qué tipo de IA se usa, qué riesgos existen y qué pasará con sus datos si se retira. Nadie midió si quienes firman entienden lo que leen, así que ese formulario no le garantiza claridad: un profesional es quien puede orientarlo.

Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504

Quién pagó: El artículo no dice quién financió el estudio ni si los financiadores tuvieron algún papel.

No tome esto como consejo médico profesional.

Los hallazgos de otros estudios que aquí se mencionan los conocemos por este documento, que fue el que leímos; no abrimos cada uno de esos estudios.

Versión detalladaLos pasajes copiados del artículo, las ilustraciones y cada fuente con cuánto leímos de ella · 1308 palabras · unos 7 minLeerla →Cerrar

Más de la mitad de los consentimientos en ensayos clínicos con IA no dicen qué IA se usa

Un análisis de 114 ensayos registrados en ClinicalTrials.gov encontró que el 58% de los formularios de consentimiento no revela el tipo de inteligencia artificial ni para qué se usa, y que el 18.4% no menciona ningún riesgo.

Si usted firmara hoy el formulario de consentimiento de uno de los 114 ensayos clínicos con inteligencia artificial que un equipo de investigadores chinos revisó en ClinicalTrials.gov, es muy probable que ese documento no le dijera qué tipo de IA se estaba usando ni para qué. Ese fue el hallazgo central: 66 de los 114 formularios, el 58%, no revelaron el tipo de IA ni su uso previsto, y 21 de ellos, el 18.4%, omitieron por completo cualquier mención de riesgos¹. El estudio fue publicado en el Journal of Medical Internet Research por Hankun Su, Fen Xiao y colegas del Hospital Xiangya de la Universidad Central del Sur, en Changsha, China, con fecha del 2 de julio de 2026. No hubo pacientes en el estudio: se analizaron documentos públicos, no personas².

¿Qué es exactamente lo que estos ensayos hacen? En la mayoría de los casos, aprendizaje automático: 39 de 92 ensayos que sí especificaron su tecnología, un 42.4%, seguidos por aprendizaje profundo, con 19 casos, y una descripción vaga de "herramientas de IA o algoritmos por escenario", con 15³. Vale la pena detenerse en esa última categoría. Quince formularios se refieren a la tecnología que va a intervenir en el cuerpo o en los datos de una persona con una frase que no describe nada. Y ese es justamente el punto que el análisis expone: un paciente puede entender qué es una pastilla, qué es una cirugía, qué es una muestra de sangre. Frente a un algoritmo, el formulario debería ser el único puente entre lo que el ensayo hace y lo que la persona acepta. El estudio midió cuántos de esos puentes existen. Encontró que más de la mitad no existen.

El análisis también revisó si esos formularios eran legibles y breves. Solo 16 de 114, un 14%, cumplían ambas condiciones a la vez: menos de 15,000 caracteres, unos diez minutos de lectura, y una puntuación de legibilidad SMOG por debajo de 13, que corresponde aproximadamente a un nivel universitario⁴. Dicho de otro modo: de cada siete formularios, seis fallan en al menos uno de los dos criterios. Y los formularios de los ensayos de mayor riesgo no eran más fáciles de entender; la correlación no fue estadísticamente significativa⁴.

Sobre los datos personales, el panorama es igual de disparejo. El derecho a retirarse estaba garantizado en todos los formularios, pero qué pasa con los datos ya recolectados cuando alguien se retira era otra cosa: 51 formularios, el 44.7%, no decían absolutamente nada al respecto; 30, el 26.3%, indicaban que los datos se destruirían; 29, el 25.4%, que los datos ya recogidos seguirían usándose; y solo 4, el 3.5%, ofrecían al participante alguna opción⁵. Solo 4 de los 114 formularios ofrecían al participante alguna opción sobre qué pasaba con su información después de irse⁵.

Aquí conviene abrir el número. Imagine que en un ensayo participan cien personas y diez se retiran a mitad de camino. Según este análisis, en casi la mitad de los casos esas diez personas no habrían encontrado en el formulario una sola línea explicando si sus datos clínicos, sus imágenes o sus muestras seguirían circulando⁵. La persona se retira del ensayo, pero no necesariamente del sistema que ya guardó su información. Esa distinción entre retirarse y desaparecer es la que el formulario debería aclarar.

Los autores proponen un marco llamado "Requisitos mínimos para el consentimiento informado en ensayos clínicos relacionados con IA", o MRIC-AI, con quince puntos que incluyen límites de extensión, puntuaciones de legibilidad, al menos una ayuda visual probada, opciones de consentimiento digital por capas, y la obligación de explicar en lenguaje llano qué datos se guardan, quién accede a ellos y qué ocurre cuando alguien se retira⁶⁷⁸. Los autores también validaron sus instrumentos con cuatro expertos, tres en medicina y uno en derecho, que valoraron cada uno de sus puntos⁶. Es una propuesta, no una solución comprobada.

El propio estudio reconoce sus límites: solo analizó documentos de consentimiento disponibles públicamente en el registro, y esos documentos pueden diferir de las versiones finales que se firman en el sitio⁹. Tampoco midió si los participantes entendieron los formularios; midió qué decían y qué tan difíciles eran de leer. Y el registro elegido, ClinicalTrials.gov, es el único que publica estos documentos¹⁰.

Ese último punto importa mucho para quien vive en América Latina, Canadá o España. Los registros de ensayos clínicos de otros países no publican los formularios de consentimiento¹⁰. Eso significa que para un ensayo registrado en Argentina, en México, en Colombia o en España, no existe forma de hacer este mismo análisis con documentos públicos. No porque los formularios no existan, sino porque nadie los publica donde cualquiera pueda revisarlos. El estudio no dice que la situación en América Latina sea igual ni peor: dice que es invisible desde afuera.

Lo que sí sugiere el contexto es que la demanda de transparencia ya está instalada. El artículo describe un estudio en el que más del 80% de los encuestados cree que debería ser notificado cada vez que se usa IA en su atención médica, sin importar el nivel de riesgo¹¹. Ocho de cada diez personas piden saber. Y según este análisis, en los ensayos clínicos con IA, más de la mitad de los formularios no lo dicen¹.

Además, los formularios fallan en proteger a quienes más protección necesitan: el 76.3% de los documentos, 87 de 114, no identificaron ninguna población vulnerable¹². Personas mayores, embarazadas, menores, comunidades con acceso limitado a la salud: ninguna de esas categorías aparece mencionada en tres de cada cuatro formularios. Y solo 13 de 114, el 11.4%, incluían alguna ayuda visual, y su presencia no se correlacionó con menor dificultad de lectura¹³.

Así lo leemos nosotros. Hay un patrón que se repite en muchos documentos que otorgan poder: quien redacta describe lo que conviene y omite lo que generaría preguntas. No se trata de un descuido de redacción. Se trata de una decisión sobre qué información merece estar en el papel y qué información no. Si ese patrón se sostiene, lo que usted puede esperar en su propia experiencia es un formulario que le habla de beneficios, que le pide la firma, y que deja fuera precisamente los detalles que usted querría discutir: qué IA es, quién la supervisa, qué pasa con sus datos si se arrepiente. Y el margen real de negociación probablemente sea mínimo.

¿Cómo sabría usted que nos equivocamos? Si un día le entregan un formulario de consentimiento para un ensayo con IA que explica en lenguaje sencillo qué tecnología se usa, quién la controla, qué riesgos existen y qué opciones tiene sobre sus datos, entonces este patrón no se cumple. Si además ese formulario le ofrece casillas para aceptar o rechazar usos concretos de su información, y le dicen qué pasa si dice no a una parte, entonces hay margen real. Eso sería la señal de que las cosas cambiaron.

Lo que usted puede hacer con esto es concreto. Antes de firmar un consentimiento para un ensayo clínico donde se mencione inteligencia artificial, pida por escrito tres cosas: qué tipo de IA se usa y para qué, quién la supervisa, y qué ocurre con sus datos si usted se retira. Pida que le expliquen oralmente lo que no entienda, tome notas o lleve a alguien de confianza. Pregunte si el consentimiento es todo o nada, o si hay opciones separadas para distintos usos de sus datos. Y si no le dan esas respuestas por escrito, eso también es información: le dice qué tipo de relación está a punto de aceptar.

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

Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - el artículo del que trata esta nota — el artículo completo — el pasaje: "More than half (66/114, 58%) of ICs failed to disclose the AI type or its intended use, and 18.4% (n=21) omitted risks entirely."
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - el artículo del que trata esta nota — el artículo completo — el pasaje: "There were no human participants in this study."
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The most commonly described AI technology in the trials was machine learning (39/92, 42.4%), followed by deep learning (n=19, 20.7%) and a general description of “AI-specific tools or scenario-based algorithms” (n=15, 16.3%; Table 1)."
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Only 14% (n=16) of ICs met the dual criteria of brevity (<15,000 characters) and readability (Simple Measure of Gobbledygook <13)."
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Data handling protocols after participant withdrawal were inconsistent: 51 (44.7%) ICs provided no information, 30 (26.3%) specified data destruction, 29 (25.4%) allowed continued use, and only 4 (3.5%) offered participants a choice."
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - el artículo del que trata esta nota — el artículo completo — el pasaje: "We propose the “Minimum Requirements for Informed Consent in AI‑related Clinical Trials” as a possible framework to improve consent quality."
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Limit document length to ≤15,000 characters (approximately 10 minutes of reading time) Achieve a Simple Measure of Gobbledygook readability score ≤13 (≤10 for high-risk trials) Include at least 1 comprehension-tested visual aid Offer layered or interactive digital consent options, if feasible"
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Specify data types, storage, access, and retention Clarify postwithdrawal data handling and offer choice Disclose future data use and consent mechanisms Cite applicable protections in plain language"
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - el artículo del que trata esta nota — el artículo completo — el pasaje: "However, it should be noted that these findings are limited to publicly available consent documents in the registry and may differ from final onsite versions."
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - el artículo del que trata esta nota — el artículo completo — el pasaje: "ClinicalTrials.gov was searched as the primary database because it is the only clinical trial registry that provides publicly available IC documents; other registries do not offer such materials"
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - el artículo del que trata esta nota — el artículo completo — el pasaje: "A recent US national survey found that more than 80% of respondents believe they should be notified whenever AI is used in their health care, regardless of risk level."
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Identification of vulnerable populations was inadequate, with 76.3% (87/114) of ICs failing to specify any such groups, despite the need for special protections."
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Only 11.4% (n=13) of ICs included visual aids, and their inclusion was not correlated with lower reading difficulty."

Quién pagó: El artículo no dice quién financió el estudio ni si los financiadores tuvieron algún papel.

No tome esto como consejo médico profesional.

Los hallazgos de otros estudios que aquí se mencionan los conocemos por este documento, que fue el que leímos; no abrimos cada uno de esos estudios.

analysis of texts · Journal of medical Internet research · the paper, 2 Jul 2026 · free

The Form You Sign for an AI Trial Often Doesn't Say What the AI Does

Researchers in China read 114 public consent documents from ClinicalTrials.gov. More than half never named the AI or its job, and 1 in 5 mentioned no risks at all.

Short version · the longer version follows, about 6 min

Ask the weeklyAI chat

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: consent documents for clinical trials that use AI
How many: 114
Where: ClinicalTrials.gov, the US registry
When: searched on June 21, 2025
Kind of study: analysis of what people did
Who did it: Xiangya Hospital, China
The limit that matters: Only forms posted on one registry were read; signed versions may differ.

What the 114 consent forms said about your data after you quit

said nothing44.7%

said data would be destroyed26.3%

said already-collected data would keep being used25.4%

let you choose3.5%

Shares of the 114 consent forms, by what they said about data handling after a participant withdraws; the study read only forms posted on ClinicalTrials.gov, and signed versions may differ.

When you are handed a consent form for a trial that uses AI, look for three things: what the AI is and does, what can go wrong, and what happens to your data if you walk away.
weeklyAI's reading

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

A consent form is the paper that turns a stranger's study into your decision. For trials that use artificial intelligence, that paper often leaves the machine out.

Researchers at Xiangya Hospital in China examined 114 clinical trials that use AI and have consent documents posted publicly on ClinicalTrials.gov, the US registry. They searched on June 21, 2025. For each form they asked what it says about the AI, about risks and benefits, how hard it is to read, and what happens to your data.

More than half, 66 of 114, never said what kind of AI was used or what it was for. In the registry entry the AI was usually described; in the form it often was not. Of the 114 forms, 21, or 18.4%, mentioned no risks at all. Privacy was the risk most often named by the forms that did.

Reading level tells its own story. Only 16 of 114 forms, 14%, were both short enough, under 15,000 characters, and simple enough to read. Trials with higher risk did not hand people easier forms.

Then there is what happens if you leave. Of 114 forms, 51, or 44.7%, said nothing about your data after withdrawal. Thirty said the data would be destroyed. Twenty-nine said already-collected data would keep being used. Only 4, or 3.5%, let you choose.

One limit matters here. The study covers ClinicalTrials.gov, the only registry that publishes consent forms. Other registries do not publish consent forms, so what their forms say is unknown. The researchers read posted documents, not the final versions signed on site, so yours may differ. And no participants were studied: the forms were measured, not whether anyone understood them.

The researchers propose a checklist for consent in AI trials, with the AI named, risks and benefits side by side, plain language and a choice about your data. It has been rated by four experts. Whether it improves anything has not been tested.

When you are handed a consent form for a trial that uses AI, look for three things: what the AI is and does, what can go wrong, and what happens to your data if you walk away.

What this means for you

If you or someone you know is handed a form for a trial that uses AI, ask the study team what the AI does, what can go wrong, and what happens to your data if you leave. This study read posted forms from one registry and did not measure whether anyone understood them, so treat it as a reason to ask questions, not as proof about trials elsewhere.

Who paid: The article does not say who funded the study or whether funders had any role.

Do not take this as professional medical advice.

The findings of other studies mentioned here are known to us through this document, which is the one we read; we did not open each of those studies.

The longer versionThe passages copied from the paper, the pictures, and every source with how much of it we read · 1233 words · about 6 minRead it →Close

The Form You Sign May Not Name the AI. A New Review of 114 Trials Found Most Don't.

Consent documents for AI trials often leave out what the AI does, what the risks are, and what happens to your data if you quit.

Researchers at Xiangya Hospital in Changsha, China, reviewed 114 clinical trials involving artificial intelligence that had consent documents posted on ClinicalTrials.gov. More than half of those documents — 66 of 114, or 58 percent — never said what kind of AI was being used or what it was for, and 18.4 percent (21 documents) did not mention risks at all¹. Only 14 percent (16 documents) were both short enough to read in about ten minutes and written simply enough for a general reader². The study, published in the Journal of Medical Internet Research, was a cross-sectional content analysis — a snapshot of documents, preregistered on the Open Science Framework and reported according to the STROBE checklist. No human participants were involved³.

The handling of data after a person quits was the least consistent part. Every consent form guaranteed the right to withdraw, but 51 documents (44.7 percent) said nothing about what would happen to data already collected. Thirty (26.3 percent) said data would be destroyed, 29 (25.4 percent) said it would keep being used, and only 4 (3.5 percent) let the participant choose⁴⁵. Just 13 documents (11.4 percent) included any picture or diagram, and having one did not make the form easier to read — the study found no correlation between visual aids and lower reading difficulty⁶. The authors themselves note the findings describe only the documents posted in the registry and may differ from the versions actually signed on site⁷.

The study is the first to look at how AI shows up in consent forms across a trial registry, according to the authors⁸. It searched ClinicalTrials.gov specifically because it is the only clinical trial registry that makes consent documents publicly available; other registries do not publish them⁹. That matters for a reader in Latin America or Canada: the picture outside that one registry is simply not known from this work. The trials that were included spanned 32 clinical domains, from cancer to cardiology to psychiatry.

What kind of AI are we talking about? In the 92 trials that named a specific technology, the most common was machine learning (39 trials, 42.4 percent), followed by deep learning (19, 20.7 percent) and general descriptions of "AI-specific tools or scenario-based algorithms" (15, 16.3 percent)¹⁰. Machine learning, in plain terms, means software that finds patterns in data rather than following rules a person wrote by hand. Deep learning is a version of that using layered networks. When a consent form omits this entirely, a participant cannot know whether the AI is reading their scan, predicting their risk, or just sorting paperwork.

Other researchers looking at AI in clinical settings have found a similar pattern of uneven disclosure. According to the summary of a study of official guidance on AI scribes — tools that listen to a consultation and draft the notes — expectations for safe use are scattered across privacy, professional, medical device, safety and health-system guidance, and while transparency is common, consent is uneven¹¹¹². We could read only the summary of that study. The same summary notes that official documents do not define what support a clinician needs at the moment of the encounter to catch a machine's invented detail, an omission, or a model that drifts over time¹³.

In psychiatry, a separate review of European rules found that AI is increasingly used for triage, risk prediction, passive monitoring and patient-facing conversation, and that European regulation addresses market access and data governance but remains fragmented on civil liability and the standard of care¹⁴¹⁵. We could read only the summary. That review also found that existing European instruments — the AI Act, the Medical Device Regulation, the data protection rules — create governance duties but do not provide a harmonized fault-based liability framework for AI-assisted care¹⁶. For a patient, the practical consequence is that a form may be legally compliant and still not tell you who is answerable if something goes wrong.

How would a reader know whether a form is doing its job? The authors propose a checklist — the "Minimum Requirements for Informed Consent in AI-related Clinical Trials" — that would require forms to name the AI type, role and level of autonomy; to state clinical, privacy and algorithmic risks with the same prominence as benefits; to stay under 15,000 characters and at a reading level a general adult can manage; to include at least one tested visual aid; and to spell out data types, storage, access, retention and what happens after withdrawal¹⁷¹⁸¹⁹. This checklist was proposed and rated by a panel of four experts — three in medicine, one in law¹⁷.

Public appetite for this information appears high. A US national survey cited in the study found more than 80 percent of respondents believe they should be told whenever AI is used in their health care, regardless of risk level²⁰. The study also found that 76.3 percent of consent documents did not identify any vulnerable populations needing special protection, despite the need for special protections²¹. Together, these two findings point to a gap between what people say they want and what the paperwork delivers.

Here is how we read it. Consent forms in this study look like documents written to satisfy a reviewer, not to be understood by the person signing. That is not a claim about anyone's motives — it is a pattern, and patterns like this usually come from systems that reward the appearance of compliance. What we expect in homes like yours: if a family member joins an AI trial, the parts of the form that matter most — what the algorithm decides, where the data goes, what is kept after you quit — will be the hardest to find and the hardest to read, and the person enrolling them will end up explaining it out loud instead. You would know we are wrong if you were handed a short, plain form that names the AI, lists the risks, and lets you choose what happens to your data.

For readers in communities that have been studied before without seeing the benefit, the missing data section is likely to land differently than it does for someone who has never had reason to ask. We would expect those readers to want limits the form does not offer. You can test whether that matters to you before you sign: ask whether you may restrict future use of your data, ask what happens if you say no, and ask for the plain-language summary and any diagram. Ask the person enrolling you to walk through the data section out loud, and keep your own written note of what you were told, with names and dates.

What this makes possible for you is a short list of questions you can carry into any clinic, any trial, any waiting room where someone hands you a form about AI. Which AI is this, and what does it decide? What are the risks you are not telling me about? If I quit, what happens to what you already have? Who do I call? If the form cannot answer, ask the research team in writing — and keep the answer.

What would you ask before you sign?

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

Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - the article this story is about — the whole article — the passage: "More than half (66/114, 58%) of ICs failed to disclose the AI type or its intended use, and 18.4% (n=21) omitted risks entirely."
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - the article this story is about — the whole article — the passage: "Only 14% (n=16) of ICs met the dual criteria of brevity (<15,000 characters) and readability (Simple Measure of Gobbledygook <13)."
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - the article this story is about — the whole article — the passage: "There were no human participants in this study."
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - the article this story is about — the whole article — the passage: "Data handling protocols after participant withdrawal were inconsistent: 51 (44.7%) ICs provided no information, 30 (26.3%) specified data destruction, 29 (25.4%) allowed continued use, and only 4 (3.5%) offered participants a choice."
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - the article this story is about — the whole article — the passage: "Although the right to withdraw was universally guaranteed in all ICs, the protocols for handling participant data after withdrawal were inconsistent."
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - the article this story is about — the whole article — the passage: "Only 11.4% (n=13) of ICs included visual aids, and their inclusion was not correlated with lower reading difficulty."
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - the article this story is about — the whole article — the passage: "However, it should be noted that these findings are limited to publicly available consent documents in the registry and may differ from final onsite versions."
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - the article this story is about — the whole article — the passage: "To date, there has been no global assessment of how AI is represented in IC documentation within clinical trial registries. This study fills that gap by conducting a cross-sectional analysis of registered clinical trials to evaluate the prevalence, clarity, and completeness of AI-related consent disclosures."
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - the article this story is about — the whole article — the passage: "ClinicalTrials.gov was searched as the primary database because it is the only clinical trial registry that provides publicly available IC documents; other registries do not offer such materials"
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - the article this story is about — the whole article — the passage: "The most commonly described AI technology in the trials was machine learning (39/92, 42.4%), followed by deep learning (n=19, 20.7%) and a general description of “AI-specific tools or scenario-based algorithms” (n=15, 16.3%; Table 1)."
Atiku SO. (2026). Regulating the drafting fiction: A qualitative content analysis of official guidance and regulator documents on AI scribe adoption and use in healthcare. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001776 — only the abstract - the full text could not be fetched — the passage: "Artificial intelligence scribes use automatic speech recognition and generative language models to document clinical consultations, but official expectations for safe adoption are scattered across privacy, professional, medical device, safety and health-system guidance."
Atiku SO. (2026). Regulating the drafting fiction: A qualitative content analysis of official guidance and regulator documents on AI scribe adoption and use in healthcare. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001776 — only the abstract - the full text could not be fetched — the passage: "Across guidelines, transparency is common, consent is uneven, verification is treated as the central safeguard in direct AI-scribe guidance, privacy governance extends from secure storage to voice identifiability, and medical device classification shifts when transcription becomes summarisation."
Atiku SO. (2026). Regulating the drafting fiction: A qualitative content analysis of official guidance and regulator documents on AI scribe adoption and use in healthcare. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001776 — only the abstract - the full text could not be fetched — the passage: "Organisational safety cases, vendor due diligence and data protection assessments narrow product-level uncertainty, but they do not define the encounter-level support needed for clinicians to detect hallucination, omission, automation bias or model drift."
Madeira L, Águas C, Schouler-Ocak M, Samochowiec J, Galderisi S, Pinto da Costa M. (2026). Liability and Standard of Care in AI-Driven Psychiatric Practice: European Viewpoint. Journal of Medical Internet Research. https://doi.org/10.2196/99894 — only the abstract - the full text could not be fetched — the passage: "AI is increasingly incorporated into psychiatric triage, risk prediction, passive monitoring, clinical documentation, and patient-facing conversational systems."
Madeira L, Águas C, Schouler-Ocak M, Samochowiec J, Galderisi S, Pinto da Costa M. (2026). Liability and Standard of Care in AI-Driven Psychiatric Practice: European Viewpoint. Journal of Medical Internet Research. https://doi.org/10.2196/99894 — only the abstract - the full text could not be fetched — the passage: "European regulation is developed in relation to market access, data governance, risk management, and product safety, yet remains fragmented regarding civil liability, organizational negligence, and the psychiatric standard of care."
Madeira L, Águas C, Schouler-Ocak M, Samochowiec J, Galderisi S, Pinto da Costa M. (2026). Liability and Standard of Care in AI-Driven Psychiatric Practice: European Viewpoint. Journal of Medical Internet Research. https://doi.org/10.2196/99894 — only the abstract - the full text could not be fetched — the passage: "Existing European instruments, including the AI Act, Medical Device Regulation, General Data Protection Regulation, revised Product Liability Directive, and European Health Data Space Regulation, establish governance duties, but do not provide a harmonized fault-based liability framework for AI-assisted health care."
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - the article this story is about — the whole article — the passage: "We propose the “Minimum Requirements for Informed Consent in AI‑related Clinical Trials” as a possible framework to improve consent quality."
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - the article this story is about — the whole article — the passage: "Limit document length to ≤15,000 characters (approximately 10 minutes of reading time) Achieve a Simple Measure of Gobbledygook readability score ≤13 (≤10 for high-risk trials) Include at least 1 comprehension-tested visual aid Offer layered or interactive digital consent options, if feasible"
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - the article this story is about — the whole article — the passage: "Specify data types, storage, access, and retention Clarify postwithdrawal data handling and offer choice Disclose future data use and consent mechanisms Cite applicable protections in plain language"
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - the article this story is about — the whole article — the passage: "A recent US national survey found that more than 80% of respondents believe they should be notified whenever AI is used in their health care, regardless of risk level."
Su, H., Xiao, F., Chau, H. et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. https://doi.org/10.2196/94504 - the article this story is about — the whole article — the passage: "Identification of vulnerable populations was inadequate, with 76.3% (87/114) of ICs failing to specify any such groups, despite the need for special protections."

Who paid: The article does not say who funded the study or whether funders had any role.

Do not take this as professional medical advice.

The findings of other studies mentioned here are known to us through this document, which is the one we read; we did not open each of those studies.