Scope of Use/Availability
  • Μany AI cough detection applications in the context of Covid-19 are available on the web, some of which are also available for android and iOS software.
  • Examples: Cambridge sounds up; Breathe for science NYU app; Coughvid by Ecole Politechnique Federeale de Lausanne; Voicemed; Detectnow; Covid Voice Deatector by Carnegie Melon University; Healthmode cough; Cough against Covid by Wadhwani Institute for AI; Virufy by Standford University; Novoic Covid-19 app; Covid cough app by Hyfe; Covid-19 Voice Study (only voice)
  • The aim of the AI cough detection programs and applications varies. Some aim only to collect data for research purposes, others provide an instant diagnosis on Covid-19.
  • A common element of all the programs and applications is that they do not aim to substitute or replace formal COVID-19 tests or medical examination.
  • Another common element is that almost all of them function on the basis of donations of cough and/or breathing/voice samples.
Technological robustness and efficacy
  • Many of the programs and applications claim to be very accurate with a percentage of more than 95% in the detection of Covid-19 cases, even of asymptomatic patients. For many programs and applications however no clinical evidence has been published supporting that they could spot the cough characteristic of Covid-19.[1]
  • During a study by Essex Univesity used 8,380 clinically-validated samples from hospitals in Spain and Mexico since April last year – 2,339 COVID-19 positive and 6,041 COVID-19 negative – the DeepCough3D screening tool proved to be 98% accurate in identifying whether the samples were positive or negative. 98% accurate in identifying whether the samples were positive or negative.[2]
  • At the same time for other applications there is no clinical proof for their efficacy to detect Covid-19. There is unknown effort to mitigate possible inaccurate predictions. For most applications, the data (recorded coughs, voice samples, etc.) were contributed by the users themselves, who make the declaration whether they are infected or not. No verification and external validation procedures were in place. No medical professionals were involved in the process of the collection, categorization and verification of the data.
  • An element of concern is the training data on which the AI cough detection applications were trained. It is unclear if they represent a diverse sample of people (age, gender, race, demographics, geographical). The model can provide inaccurate results for people whose characteristics are underrepresented in the trials.
  • In general, more research in this area is required to assess the actual efficacy of using AI for cough detection.

[1] MIT algorithm


Impact on citizens and society (blue=pos, red=neg, orange=pos+neg)
  • This type of AI use could have a negative impact on citizens and society, if these applications would be used widely throughout society, rather than in a confined medical setting.
  • Wide use could set an undesired precedent for the future (more (acceptance of)) surveillance, with ample impact on the human right to a private life.
  • Compliance with the GDPR can be difficult when used widely, as it deals with sensitive personal data (biometric data).
  • There is impact on human agency and free will where use of the system could become mandatory or ‘hidden’.
  • If clinically validated and used in clinical settings it could provide a low cost early indicator of COVID-19 infections, which could help in triage.
  • The functionality and effectiveness of this algorithm and its further optimization in the context of Covid-19 could be promising and useful for the detection and recognition of other diseases in the future.[1]


Governance and Accountability
  • Processing of personal data conducted by AI cough detection programs and applications in the context of Covid-19 through voluntary provision of coughs and voice samples is in compliance with the requirement for a legal basis for processing prescribed in the GDPR provided that the individual has given their explicit and free consent and the data is not used for other purposes.
  • It could be claimed that other legal bases for processing of personal data are the interest of other natural persons and the public interest, including specifically public interest for health reasons in case of special categories of personal data, i.e. combat of Covid-19 pandemic.
  • Scientific research benefits of a broader interpretation of the concept of consent. Thus the consent given covers the processing for additional Covid-19 related research purposes in addition to the specific research from the specific university.
  • The provision of encryption of data and contractual arrangements as measures deployed to ensure confidentiality and security of the data are important.
  • Risks arise in case of compulsory use of AI applications for cough detection for example by employees, as well as in case of wide but covert use in public spaces.
  • Especially for the latter, there is no governance structure in place.
Acceptable trade-offs in times of crisis

The use of AI cough detection programs and applications could be justified to contribute to the fight against Covid-19, however, certain conditions must be met in order to ensure their responsible use. Wider (covert) application in public or private spaces render a number of legal, ethical and governance issues. A major trade-off is that the AI applications for cough detection that are available have no sufficiently proven efficacy and robustness. Given the above we feel that at this point there is no acceptable trade-offs for the wide (indiscriminate) use of these types of applications.