Voice app startups collect data in poorer countries with no shared rules.

US · CanadaAmérica Latina

Weekly AI Watch

Every week, what changed and what it means for you. No hurry, no alarm.

Week of 23 October 2026EN | ESPast editions18,276 papers found · 17,064 read · 5,349 belonged · 37 stories passed every check

Observatorio de la
Inteligencia Artificial

Cada semana, qué cambió y qué significa para usted. Sin prisa y sin alarma.

Semana del 23 de octubre de 2026EN | ESEdiciones anteriores18,276 estudios encontrados · 17,064 leídos · 5,349 correspondían · 37 notas pasaron todas las comprobaciones

weeklyAI · Week of 23 October 2026weeklyAI · Semana del 23 de octubre de 2026

← Your rights← Sus derechos

interviews · Frontiers in digital health · la publicación, 15 sep 2026 · gratis

"No, éramos una empresa": lo que once empresas de voz e inteligencia artificial contaron sobre la revisión ética

Entrevistadas por investigadores, la mayoría describió comités de ética y formularios de consentimiento; dos no tenían revisión propia. Sólo 11 de 27 empresas contestaron, y cada relato es el de la propia empresa.

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: empresas emergentes que usan la voz como marcador de enfermedades
Cuántos: once de 27 empresas contactadas aceptaron (41%)
Dónde: Estados Unidos, Canadá, Reino Unido, Luxemburgo, Alemania, Francia, Australia y Argentina
Cuándo: entrevistas entre febrero y mayo de 2025
Tipo de estudio: interviews
Quién lo hizo: Weill Cornell Medical College, Nueva York
El límite que importa: Son las palabras de las empresas sobre sí mismas, no registros verificados; las más cuidadosas quizá aceptaron.

Dónde se recogieron las voces y dónde se pensaba vender el producto

Cuatro de las once empresas recogían datos en países de ingresos bajos y mediosfrente aSus productos iban destinados sobre todo a mercados de países ricos

Se graba donde la compensación es más baja y se vende donde están los compradores

La voz, dijeron los propios entrevistados, no se puede anonimizar del todo.
Lectura de weeklyAI

Ver · 0:31

"No, éramos una empresa. Así que, en realidad, íbamos a tener un comité de ética por medio de una colaboración con una universidad."

Así contestó el representante de una empresa emergente cuando le preguntaron por la revisión ética. Ese comité es el grupo que revisa una investigación con personas antes de que empiece. Esta empresa pensaba conseguir esa revisión después, a través de una universidad. Otra dijo que sus integrantes en instituciones académicas "no requieren esa información", y por eso no la buscó.

Los investigadores entrevistaron a representantes de 11 empresas entre febrero y mayo de 2025. Las empresas trabajan en inteligencia artificial que lee la voz para detectar o seguir padecimientos como el párkinson, el alzhéimer y la depresión. Les preguntaron cómo recogen las voces, cómo piden el consentimiento y cuánto pagan.

Dos de las 11 no habían buscado ninguna revisión ética propia. De las nueve que sí, tres usaron comités comerciales. El consentimiento variaba. Una empresa usaba casillas una por una, otra un formulario de una página, y otra dijo que a veces conseguía una dispensa. Cuatro de las 11 grababan voces en países pobres, tres de ellas en la India, mientras apuntaban sus productos a mercados ricos. Una describió el pago allá como "unos cuantos dólares al día", frente a mucho más en Estados Unidos.

Piense en lo que es una voz. Una grabación de usted no se puede limpiar de usted. Un entrevistado lo dijo sin rodeos: quitarle la identidad a la voz es "muy problemático", porque al cambiar la voz deja de servir. Otro dijo que el robo de una base de voces sería un incidente grave, porque las voces son difíciles o quizá imposibles de anonimizar. El estudio tampoco encontró una norma común para grabar o etiquetar las voces, así que los datos de una empresa no se pueden cotejar con los de otra.

El estudio no puede decirle qué permite la ley de su país. Sólo 11 de 27 empresas contestaron, y se las buscó a través de una red académica que atrae a las que ya hablan con especialistas en ética, así que las peores prácticas probablemente quedaron fuera de la cuenta. Cada relato es el de la propia empresa, sin verificar. Ninguna aparece con nombre.

Si una clínica, un empleador o una aplicación le pide la voz, las protecciones que usted da por sentadas pueden no existir, y la grabación puede durar más que la empresa que la tomó.

Busque cuatro cosas: quién pide la voz, si un comité de ética con nombre lo aprobó, qué dice el formulario sobre borrar la grabación, y cuánto le pagan.

Qué significa para usted

Cuando una clínica, una aplicación o su trabajo le pidan la voz, pregunte quién la recibe y qué dice el formulario sobre borrarla después. El estudio no revisó leyes de ningún país ni midió si esos productos funcionan, así que no concluya todavía que su voz está protegida por lo que aquí se describe.

Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383

Quién pagó: El artículo indica que se recibió apoyo financiero y que AR contó con una beca Paul B. Beeson Emerging Leaders Career Development Award in Aging (K76 AG079040) del National Institute on Aging y con el premio Bridge2AI (OT2 OD032720) del NIH Common Fund; no dice si los financiadores tuvieron alguna influencia ni menciona equipos o programas cedidos para el estudio.

No tome esto como consejo médico profesional.

Versión detalladaLos pasajes copiados del artículo, las ilustraciones y cada fuente con cuánto leímos de ella · 1449 palabras · unos 7 minLeerla →Cerrar

Graban su voz en India y África para vender salud en países ricos: lo que once empresas contaron de sí mismas

Una investigación entrevistó a empresas que usan la voz como marcador de enfermedades. El estudio no midió si sus productos funcionan; lo que salió a la luz es cómo consiguen los datos.

Un teléfono graba una frase, un programa la analiza y devuelve una probabilidad: así se presenta la voz como herramienta para detectar el párkinson, el alzhéimer, la depresión o una insuficiencia cardiaca. La promesa es enorme y no requiere agujas. El estudio que nos ocupa no midió esa promesa. Entrevistó a once empresas emergentes que trabajan en ese terreno para preguntarles de dónde salen las voces con las que entrenan sus programas, quién revisa esos estudios, qué firma la gente y qué pasa con las grabaciones después. Las respuestas están publicadas en Frontiers in Digital Health, con fecha del 15 de septiembre de 2026, y las entrevistas se hicieron entre febrero y mayo de 2025. Los autores son Yuhan, Swain, Zhao, Ramos y Rameau, del instituto de la voz del Weill Cornell Medical College, en Nueva York. Una de las autoras recibió apoyo del Instituto Nacional del Envejecimiento de Estados Unidos y de un fondo común de los Institutos Nacionales de Salud. Una de las autoras asesora a Sound Health Systems y a Pentax Medical, y fundó VAST Health.

De 27 compañías contactadas, once aceptaron hablar: el 41 por ciento. Están en Estados Unidos (tres), Canadá (dos), Reino Unido, Luxemburgo, Alemania, Francia, Australia y Argentina. Las entrevistas fueron por videollamada y duraron menos de media hora. El equipo encontró seis patrones. No hay normas compartidas para grabar: un teléfono Samsung y otro graban distinto, una computadora y un celular también, y eso mete ruido en la señal. La revisión ética va de comités universitarios formales a nada. Dos de las once empresas no habían buscado ninguna revisión propia. El consentimiento va de casillas detalladas a un consentimiento implícito y avisos mínimos, y una empresa contó que a veces obtenía una dispensa del consentimiento. La voz, dijeron los propios entrevistados, no se puede anonimizar del todo. Cuatro de las once recogían datos en países de ingresos bajos y medios, tres de ellos en India, mientras apuntaban a vender en países ricos. Y siete ya habían desarrollado productos comerciales con esos datos, y tres más estaban en proceso.

Piense en el consultorio donde le piden una muestra de sangre. Antes de que le saquen nada, alguien le explica para qué es, quién la guarda y cuánto tiempo. Esa cadena de reglas tardó décadas en construirse. Con la voz no existe equivalente, y el artículo lo dice con esas palabras al comparar ambos mundos. Lo curioso es que la voz tiene una ventaja y un problema a la vez: sirve para diagnosticar y sirve para identificar a la persona. Una empresa lo explicó sin rodeos: no se puede alterar la voz para esconder al hablante, porque entonces deja de servir.¹ Por eso el robo de una base de grabaciones sería grave de un modo distinto al de una lista de correos.

Los números de una de las empresas dejan ver el tamaño del traslado. Reportó alrededor de 50 mil participantes de investigación en India y 10 mil en cuatro países africanos, contra menos de mil en Europa.² La compensación sigue la misma línea: "En India son unos cuantos dólares al día; en Estados Unidos sería mucho más", dijo un representante.³ Los propios entrevistados describieron el incentivo de recoger datos donde sale más barato.⁴ Y el producto final no se queda ahí: iba destinado sobre todo a mercados de países ricos.⁵ El artículo nombra ese patrón con una idea prestada del estudio de datos agrícolas: la extracción de recursos biológicos de países con protección más débil para vender lo obtenido en los mercados de altos ingresos.⁶

Sobre la revisión ética, el abanico es amplio. Dos de las once no habían buscado ninguna revisión propia.⁷ Una se describió como "puramente privada", sin vínculo con universidad alguna, aunque en realidad sí había obtenido el aval de un comité nacional para su estudio clínico.⁸ Otra contó que, para no alargar el documento, buscó un punto medio entre explicar bien y ser breve.⁹ El consentimiento, cuando existe, puede ser una casilla por cada uso, o una firma en papel en Kenia, o un formulario de una sola página. Todo eso convive en el mismo sector.

Así lo leemos nosotros. El patrón es este: la voz se recoge donde hay menos instituciones que la vigilen y se convierte en producto donde hay compradores. De ahí esperamos una escena que usted podría reconocer: una comunidad que recibe la grabación como un servicio moderno, sin comité local visible que la acompañe, y una empresa que responde desde otro país. Podríamos estar equivocados: si esas comunidades tuvieran comités, autoridades o grupos de pacientes que revisan y negocian antes de la grabación, o si rechazaran participar, el patrón no se sostendría. Usted puede averiguarlo con dos preguntas antes de prestar su voz a cualquier aplicación de salud: quién responde aquí si algo sale mal, y qué autoridad de su país revisó ese estudio. Si la respuesta es "nadie" o "no sé", ya tiene su dato.

También vale la pena fijarse en el destino del archivo. Es probable que la grabación quede en servidores del Norte global y que el control se ejerza desde allí, incluso si la empresa quiebra, porque no hay reglas que obliguen a devolverla o destruirla. Sabríamos que nos equivocamos si las comunidades conservaran copia y voz en las decisiones sobre ese archivo, o si existieran contratos que reconozcan la voz como dato con valor económico y reparto de beneficios. Mientras tanto, pida por escrito qué pasa con su grabación si la empresa cierra o se vende, y qué recibe su comunidad a cambio. Si la única respuesta es "ayudar a la ciencia", téngalo en cuenta al decidir.

El estudio tiene límites que conviene decir en voz alta. Las empresas se reclutaron a través de una red académica, la del consorcio Bridge2AI, que busca justamente fijar estándares en este campo; es probable que eso haya atraído a las más cuidadosas y que las peores prácticas queden fuera de la foto.¹⁰ Once de 27 aceptaron, y los autores no pudieron caracterizar a quienes no respondieron.¹¹ Lo que se recogió son las palabras de las empresas sobre sí mismas, no registros verificados, y la conversación cara a cara invita a presentarse bien. Varias autoras pertenecen al consorcio que quiere normar el sector. El estudio no nombra a ninguna compañía y no midió si algún producto funciona o dañó a un paciente. Tampoco dice qué exige la ley de su país: eso tendría que consultarlo usted.

Dos cosas más, de otros trabajos que leímos. En un análisis de documentos de consentimiento para investigaciones con inteligencia artificial, más de la mitad no explicaba qué tipo de sistema se usaba ni para qué, y una parte no mencionaba riesgo alguno; pudimos leer solo el resumen.¹² Y en un estudio sobre datos médicos aparentemente anonimizados, se advierte que los programas pueden deducir información sensible, como la susceptibilidad a una enfermedad, a partir de datos que parecían limpios; también leímos solo el resumen.¹³

¿Qué haría falta para enderezar esto? Los autores proponen tres cosas concretas. Para las empresas: registrar metadatos mínimos en el momento de grabar, con aparato, micrófono, ambiente y cómo se asignó el diagnóstico. Para el sector: adoptar definiciones comunes y construir infraestructura compartida de datos. Para reguladores y financiadores: revisar cómo se clasifica la voz en marcos como HIPAA y el reglamento europeo, y exigir planes sobre qué pasa con las grabaciones antes de que una empresa cierre.¹⁴¹⁵ El artículo recuerda además que la voz ya figura entre los identificadores protegidos por la ley estadounidense de portabilidad de seguros de salud.

Hay una tensión que el propio artículo señala con honestidad: el riesgo de reidentificación, real, sirve también de excusa para no compartir datos.¹⁶ Y no compartir impide que alguien independiente verifique si el programa funciona. Los datos son el activo principal de estas empresas.¹⁷ Compartir expone; no compartir protege el negocio. Ambas cosas son verdad al mismo tiempo.

Vuelva al teléfono que graba una frase. Ahora sabe que esa frase puede viajar lejos, quedar sin dueño claro y alimentar un producto que quizá nunca se venda en su país. Si le ofrecen participar en algo así, pida el documento de consentimiento, busque la parte sobre qué pasa si la empresa cierra, y pregunte qué comité revisó el estudio. Con esas tres respuestas en la mano, usted decide.

*¿Qué le preguntaría usted a la empresa antes de prestar su voz?*

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

Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Representatives described voice as inherently identifiable and pointed to the difficulty of stripping identity from it: “The de-identification of voice is very problematic… I can't change the voice because… then it's not useful” (Q30)."
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - el artículo del que trata esta nota — el artículo completo — el pasaje: "For one startup the asymmetry was also quantitative: it reported approximately 50,000 research participants in India and 10,000 across four African countries, against fewer than 1,000 in Europe."
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Startups noted that compensation in these settings is lower than it would be in high-income countries (HICs): “In India, it's a few dollars a day. In the U.S., it would be a lot more” (Q41)."
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - el artículo del que trata esta nota — el artículo completo — el pasaje: "By the startups' own account, the lower compensation standards in LMICs reflect a structural incentive to collect data wherever it is cheapest."
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Yet while these companies sourced data in LMICs, their products were intended primarily for HIC markets (Q39–Q40)."
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - el artículo del que trata esta nota — el artículo completo — el pasaje: "This pattern, extracting biological data from countries with weaker regulatory protection and commercializing the resulting products in high-income markets, exemplifies the asymmetry at the heart of what Leonelli terms the epistemic economy of transnational data"
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - el artículo del que trata esta nota — el artículo completo — el pasaje: "At the other end, two of the eleven startups had sought no ethics review of their own"
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - el artículo del que trata esta nota — el artículo completo — el pasaje: "A further organization described itself as “purely private” with “no connection with any public institution or university” (Q17), yet had in fact obtained approval from a national research ethics committee for its clinical study"
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - el artículo del que trata esta nota — el artículo completo — el pasaje: "One startup revealed a potentially concerning attitude toward consent, explaining: “We wanted to make it short, and the data protection officer wanted to make everything long… we tried to make a compromise between explaining well to participants but also having it concise” (Q24)."
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Second, there is potential for selection bias: recruitment through the Bridge2AI-Voice Consortium's symposium networking list likely favored companies already engaged with the academic and ethics community, that is, those most likely to have considered IRB review, consent, and data provenance."
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Third, eleven of the 27 startups we contacted agreed to participate (41%) and we were unable to systematically characterize non-responders, and companies that declined may differ from participants in their data governance practices."
Su H, Xiao F, Chau H, Tong Y, Han S, Cheng X, et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. 10.2196/94504 — solo el resumen - no se pudo obtener el texto 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."
Jung M. (2026). Protecting Informational Self-determination in AI-driven Precision Oncology: Privacy, Ethics, and Governance Challenges. Journal of Preventive Medicine and Public Health. 10.3961/jpmph.26.094 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "However, AI systems generate risks not only through re-identification but also through inferential analytics that derive new privacy-relevant information, such as disease susceptibility or familial risk, from ostensibly anonymized data."
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - el artículo del que trata esta nota — el artículo completo — el pasaje: "For regulators and funders, we recommend reassessing the risk classification of voice under HIPAA- and GDPR-style frameworks to reflect its non-anonymizability (7, 8) and attaching benefit-sharing and community-engagement expectations to research and products built on data sourced from low- and middle-income settings (theme 5)."
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - el artículo del que trata esta nota — el artículo completo — el pasaje: "We also encourage regulators to require data-disposition plans for biometric datasets, so that the fate of voice recordings is defined before, not after, a company ceases operations."
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - el artículo del que trata esta nota — el artículo completo — el pasaje: "This quote captures how a legitimate concern (the non-anonymizability discussed above) and a competitive concern (being “overtaken”) are entangled: privacy risk supplies a defensible rationale for withholding data, while financial and reputational interests supply the underlying motive."
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Where academic funders and journals increasingly mandate data sharing, startups face negative incentives to share, because datasets are their principal corporate asset (theme 6)."

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.

interviews · Frontiers in digital health · the paper, 15 Sep 2026 · free

Voice app startups collect data in poorer countries with no shared rules.

Interviews with 11 voice-biomarker startups found ethics review, consent and pay ranging widely, from item-by-item opt-in to none at all. Only 11 of 27 companies answered, and each account is the company's own.

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: representatives of voice-biomarker startups
How many: eleven startups
Where: worldwide
When: February to May 2025
Kind of study: interviews
Who did it: Weill Cornell Medical College
The limit that matters: Only 11 of 27 companies answered, recruited through an academic network, so worse practices are probably undercounted.

Research participants one startup reported, by region

India50000people

four African countries10000people

Europe1000people

Participants one startup said it had recorded in each region; the Europe figure is given only as fewer than 1,000, and these are one company's own account, not verified.

Where voices were collected versus where products were aimed

startups collecting data in low- and middle-income countriesagainststartups aiming products at high-income markets

Four of the eleven collected in poorer countries, three of them in India, while products were intended mainly for rich markets.

If a clinic, employer or app asks for your voice, the safeguards you assume may not exist, and the recording may outlive the company that took it.
weeklyAI's reading

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

"No, we were a business. So actually, we were going to have an IRB with a collaboration with a university."

That is a startup representative, asked about ethics review. IRB means an institutional review board, a committee that checks research on people before it starts. This company planned to get that check later, through a university partnership. Another said its team members at academic institutions "do not require that information," so it did not seek review.

Researchers interviewed representatives from 11 startups between February and May 2025. The companies work on artificial intelligence that reads voices to screen for or monitor conditions such as Parkinson's, Alzheimer's and depression. The researchers asked how the companies collect voices, get consent and pay people.

Two of the 11 had sought no ethics review of their own. Among the nine that did, three used commercial review boards. Consent practices differed. One company used item-by-item checkboxes, one used a one-page form, and one said it sometimes got a waiver of consent. Four of the 11 collected voices in poorer countries, three of them in India, while aiming their products at rich markets. One described pay there as "a few dollars a day," against much more in the United States.

Consider what a voice is. A recording of you cannot be scrubbed of you. One interviewee put it plainly: de-identifying voice is "very problematic," because changing the voice makes it useless. One interviewee said a hacked voice database would be a serious incident, because the voices are hard or maybe impossible to anonymize. The study also found no shared standard for how voices are recorded or labeled, so one company's data cannot be checked against another's.

The study cannot tell you what your own country's law allows. Only 11 of 27 companies answered, and they were recruited through an academic network that favors companies already talking to ethicists, so worse practices are probably undercounted. Every account is the company's own, unverified. No company is named.

If a clinic, employer or app asks for your voice, the safeguards you assume may not exist, and the recording may outlive the company that took it.

Look for four things: who is asking, whether a named ethics board approved it, what the consent form says about deleting your recording, and what you are paid.

What this means for you

For now, the thing to hold on to is that this study describes 11 companies and cannot say how common any of it is. If an app or an employer ever asks for a voice sample, the useful question is the plainest one: who holds the recording, and what happens to it if the company closes. Ask that before you speak.

Who paid: The article says financial support was received and that AR was supported by a Paul B. Beeson Emerging Leaders Career Development Award in Aging (K76 AG079040) from the National Institute on Aging and by the Bridge2AI award (OT2 OD032720) from the NIH Common Fund; it does not say whether the funders had any say, and names no equipment or software lent to the study.

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 · 1243 words · about 6 minRead it →Close

Two of Eleven Voice Startups Sought No Ethics Review at All

A study of 11 voice-biomarker companies found no shared data standards, ethics review that ranged from thorough to absent, and recordings gathered where pay is lowest for products aimed at the richest markets.

Eleven companies are trying to turn your voice into a medical test. The pitch is that software can hear conditions such as Parkinson's disease, Alzheimer's disease, depression or heart failure in the sound. Researchers at Weill Cornell Medical College interviewed one representative from each of eleven such startups, chosen from 27 they approached, and published the results in Frontiers in Digital Health. The interviews ran from February to May 2025 and lasted under half an hour each. No patient was examined, no product was tested, and no one's health outcome was measured.

Consider what a recording actually contains. The article describes a study in which a voice carries the disease signal a company wants and the identity of the person who spoke, and the two cannot be separated: strip out what makes a voice recognizable and the medical signal goes with it¹. That single property drives much of what the interviews describe. The article describes a study in which, because a voice cannot truly be made anonymous, a leaked recording is a lasting exposure, and companies know it¹. It also gives them a reason, or an excuse, to keep recordings locked away. The article describes a study in which the authors put it plainly: the privacy worry is real, and the competitive worry rides along with it, since a firm that shares its data risks being overtaken².

Six patterns emerged. There were no shared standards for how to record or label a voice sample. Ethics review ran from full multi-site approval to nothing at all: two of the eleven companies had sought no ethics review of their own³. Consent ranged from item-by-item opt-in to a one-page form. Voice was acknowledged as impossible to anonymize. Four of the eleven collected data in low- and middle-income countries, three of them in India⁴, while aiming their products at rich markets⁵. And most kept their data proprietary.

The scale of one company's sourcing is worth sitting with. A single startup told the researchers it had about 50,000 participants in India and 10,000 across four African countries, against fewer than 1,000 in Europe⁶. Picture a stadium filled in one country and a mid-sized town's worth of people in four others, all to build a product for a third market.

The pay gap was stated by the companies themselves. One put it as "a few dollars a day" in India against "a lot more" in the United States⁷. The authors read the pattern as a structural pull toward wherever collection is cheapest⁸. Grant the companies this much: a lower payment in a poorer country buys more there, and a small sum is not automatically an unfair one. The study did not measure whether anyone was underpaid. The article describes a study in which what it records is the direction of the flow: voices gathered where oversight and pay are lower, products sold where they are higher⁹.

It is worth comparing the two ends of the ethics spectrum. On one side, a company described standard national review plus a commercial central board used across American sites. On the other, two of the eleven companies had sought no ethics review of their own³. In between sat a firm that called itself "purely private" with no university tie — and had in fact obtained approval from a national research ethics committee¹⁰. The system was there. Some used it, some did not, and one described itself as outside it while standing inside.

Consent showed the same spread. One company compressed its disclosure to a one-page form, balancing its wish for brevity against a data-protection officer's wish for length¹¹. Another offered granular boxes, one per use. The study's authors note that consent is meant to be an ongoing, understood process, not a formality — a standard a deliberately short form may fail.

Here is how we read it. The pattern is familiar from other new industries: the public face of a company is its promise of universal health benefit, while the arrangements that decide who holds the recordings, who profits, and whether anyone ever reviewed the study stay out of sight. We expect that if you ask a voice-health company today which ethics committee approved its work, in which countries people were recorded, what they were paid and who owns the files, you will often get a brochure and not an answer. We could be wrong: we would be wrong if companies began publishing those four facts in plain language before collection starts. Until then, the useful move is to ask in writing, to the company and to your country's data-protection authority, and to keep the reply.

We also read a second pattern, and it concerns you directly if you are ever asked to record your voice. The benefit in front of a participant is visible; what is given up, including uses that appear years later, is not. We expect that consent for voice collection in lower-income settings will often be short, delegated to a local partner, or skipped, and that participants will rarely be told their recordings may end up in products sold elsewhere. We would be wrong if we found consent explained in the local language, with future uses and re-identification risk spelled out, and a complaint route that works. So before you tap agree, ask what happens to your recording if the company closes, and ask which authority you can complain to.

The study has limits that matter. Only eleven of 27 companies answered, and they were recruited through an academic network, which likely favored firms already thinking about ethics¹². The companies that declined could not be characterized, and may differ¹³. The interviews record what companies said about themselves, not verified records. The study names no company and reports no country's law. Several authors belong to a consortium that promotes data standards in this very field. And the transcripts cannot be released, so no outsider can check the quotes.

The authors' own proposals are concrete. The article describes a study in which they ask regulators to reassess how voice is classified under privacy laws such as HIPAA and GDPR, given that it cannot be anonymized¹⁴, to attach benefit-sharing and community-engagement conditions to data gathered in poorer countries¹⁴, and the article describes a study in which they ask to require companies to say in advance what happens to recordings when they shut down¹⁵. That last point is not hypothetical: one company in the sample had already halted after a failed fundraising round, holding psychiatric recordings with no stated plan for them.

What this makes possible for you is a short list of questions, asked early. Which ethics board approved this, and can I see its name? In which countries were people recorded, and what were they paid? Who owns my recording, and where is it stored? What happens to it if the company folds? A company that answers those four in writing has told you something a brochure cannot.

The voice that a company wants for its test is the same voice that identifies you at a bank, on a phone call, in a courtroom. Ask who holds it before you lend it: that question is yours to put, and no ethics board will put it for you.

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

Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "Representatives described voice as inherently identifiable and pointed to the difficulty of stripping identity from it: “The de-identification of voice is very problematic… I can't change the voice because… then it's not useful” (Q30)."
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "This quote captures how a legitimate concern (the non-anonymizability discussed above) and a competitive concern (being “overtaken”) are entangled: privacy risk supplies a defensible rationale for withholding data, while financial and reputational interests supply the underlying motive."
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "At the other end, two of the eleven startups had sought no ethics review of their own"
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "Four of the eleven startups collected data from low- and middle-income countries (LMICs), three of them in India."
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "Yet while these companies sourced data in LMICs, their products were intended primarily for HIC markets (Q39–Q40)."
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "For one startup the asymmetry was also quantitative: it reported approximately 50,000 research participants in India and 10,000 across four African countries, against fewer than 1,000 in Europe."
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "Startups noted that compensation in these settings is lower than it would be in high-income countries (HICs): “In India, it's a few dollars a day. In the U.S., it would be a lot more” (Q41)."
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "By the startups' own account, the lower compensation standards in LMICs reflect a structural incentive to collect data wherever it is cheapest."
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "This pattern, extracting biological data from countries with weaker regulatory protection and commercializing the resulting products in high-income markets, exemplifies the asymmetry at the heart of what Leonelli terms the epistemic economy of transnational data"
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "A further organization described itself as “purely private” with “no connection with any public institution or university” (Q17), yet had in fact obtained approval from a national research ethics committee for its clinical study"
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "One startup revealed a potentially concerning attitude toward consent, explaining: “We wanted to make it short, and the data protection officer wanted to make everything long… we tried to make a compromise between explaining well to participants but also having it concise” (Q24)."
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "Second, there is potential for selection bias: recruitment through the Bridge2AI-Voice Consortium's symposium networking list likely favored companies already engaged with the academic and ethics community, that is, those most likely to have considered IRB review, consent, and data provenance."
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "Third, eleven of the 27 startups we contacted agreed to participate (41%) and we were unable to systematically characterize non-responders, and companies that declined may differ from participants in their data governance practices."
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "For regulators and funders, we recommend reassessing the risk classification of voice under HIPAA- and GDPR-style frameworks to reflect its non-anonymizability (7, 8) and attaching benefit-sharing and community-engagement expectations to research and products built on data sourced from low- and middle-income settings (theme 5)."
Yuhan, A., Swain, S., Zhao, R. et al. (2026). The epistemic economy of vocal biomarker startups: a qualitative interview study of data practices and governance. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1895383 - the article this story is about — the whole article — the passage: "We also encourage regulators to require data-disposition plans for biometric datasets, so that the fate of voice recordings is defined before, not after, a company ceases operations."

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 free weekly drip by email

One email a week: the free story of every section, with its picture, and the headlines of the rest.

No payment, nothing to cancel. Unsubscribe from any issue.

El goteo semanal gratis por correo

Un correo por semana: la nota gratuita de cada sección, con su imagen, y los titulares de las demás.

Sin pago y sin nada que cancelar. Puede darse de baja desde cualquier envío.