New · France 2027 presidential election: what the candidates propose on AI, quoted and sourced. Explore the tracker →

Last reviewed:

What is trustworthy AI? Definition, the 7 requirements and what an executive can demand

Trustworthy AI is AI that is lawful, ethical and robust and that meets seven requirements set out by the European Commission's High-Level Expert Group on 8 April 2019: human agency and oversight, technical robustness and safety, privacy and data governance, transparency, diversity and fairness, societal and environmental well-being, accountability. The AI Act takes up its spirit.

The expression comes from the Ethics Guidelines for Trustworthy AI, published on 8 April 2019 by the High-Level Expert Group on AI appointed by the European Commission. Trustworthy AI must be lawful (compliant with law), ethical (compliant with values) and robust (technically and socially). It meets seven requirements: human agency and oversight; technical robustness and safety; privacy and data governance; transparency; diversity, non-discrimination and fairness; societal and environmental well-being; accountability. The text is not binding. It comes with an assessment checklist, ALTAI, published on 17 July 2020 after a pilot with more than 350 participants. The AI Act (Regulation (EU) 2024/1689) made it a guiding thread: Article 1 aims to promote human-centric and trustworthy AI, and recital 27 takes up the seven principles as a non-binding basis. In France, the Confiance.ai programme, launched in 2021 under a State Grand Challenge and led by IRT SystemX, brought together thirteen structuring partners (Airbus, Thales, Renault, Safran, Valeo, CEA, Inria and others) around a budget of 45 million euros over four years, to industrialise trustworthy AI in critical systems. It ended in September 2024; a European foundation was announced to carry its tools forward. Do not confuse it with responsible AI, which concerns implementation and accountability: see the table.

Concrete example

Illustrative case: a 400-employee equipment maker chooses an AI supplier for the visual inspection of its welds. Rather than relying on the sales brochure, the quality department asks the supplier to document each of the seven requirements: who supervises decisions and can correct them, how robustness was tested (light, dust, new parts), where the training data comes from, how a decision is explained to an operator, what performance gaps exist across production lines, and who is responsible in case of error. Two answers are judged insufficient: the supplier completes them before signing. The file also becomes the basis for the annual contract review.

Comparison

Trustworthy AI, responsible AI, ethical AI: three expressions, three angles
Trustworthy AIResponsible AIEthical AI
Main referenceGuidelines of the Commission's expert group (2019)OECD AI Principles (2019, revised 2024)Values and charters specific to each organisation
ContentThree components (lawful, ethical, robust), seven requirementsFive principles, focused on implementationNo common list
Question askedCan this system be relied upon?Who answers for what, with what evidence?Is what we do right?
Legal statusNon-binding; spirit taken up by the AI ActNon-binding; ISO 42001 for the organisationNone
Associated toolALTAI (2020)Register of uses, ISO/IEC 42001Charter, ethics committee

FAQ

What is trustworthy AI?

Trustworthy AI is AI that is lawful, ethical and robust and that meets seven requirements set by the European Commission's High-Level Expert Group in April 2019. It is a reference for the AI Act, but the term is not defined as such in the law.

What is the definition of trustworthy AI?

According to the 2019 European guidelines, it is AI that is lawful, ethical and robust, and that meets seven requirements: human agency and oversight, robustness and safety, privacy and data, transparency, diversity and fairness, societal and environmental well-being, accountability.

What are the 7 requirements of trustworthy AI?

1. Human agency and oversight. 2. Technical robustness and safety. 3. Privacy and data governance. 4. Transparency. 5. Diversity, non-discrimination and fairness. 6. Societal and environmental well-being. 7. Accountability.

What is the difference between trustworthy AI and responsible AI?

Trustworthy AI rests on a specific European framework: three components and seven requirements. Responsible AI refers rather to the organisation's approach (OECD principles, governance, evidence) to design, buy and use AI. The first describes the expected result, the second the way to get there.

Is trustworthy AI mandatory?

The seven 2019 requirements are non-binding. The AI Act takes up their spirit and makes part of it mandatory depending on the risk level: prohibitions, transparency, requirements for high-risk systems. Outside these cases the approach remains voluntary.

What is Confiance.ai?

It is a French research and engineering programme launched in 2021, led by IRT SystemX, funded by France 2030 (45 million euros over four years at launch) and bringing together industrial groups and laboratories. It targeted trustworthy AI in critical systems and ended in September 2024.

How can you tell whether an AI is trustworthy?

No single label guarantees it. You ask for evidence on the seven requirements: human oversight, robustness tests, data origin, explanations, bias measurement, a named owner. The Commission's ALTAI checklist (2020) provides a self-assessment framework.

What is ALTAI?

ALTAI (Assessment List for Trustworthy AI) is a self-assessment checklist published on 17 July 2020 by the High-Level Expert Group. It turns the seven requirements into concrete questions for AI developers and deployers.

See also

Further reading

Ethics Guidelines for Trustworthy AI, European Commission, 8 April 2019 (external resource)

Sources

  1. Ethics Guidelines for Trustworthy AI, High-Level Expert Group on AI, European Commission, 8 April 2019. https://digital-strategy.ec.europa.eu/en/library/ethics-guidelines-trustworthy-ai (accessed 2026-11-14)
  2. Assessment List for Trustworthy AI (ALTAI), European Commission, 17 July 2020. https://digital-strategy.ec.europa.eu/en/library/assessment-list-trustworthy-artificial-intelligence-altai-self-assessment (accessed 2026-11-14)
  3. Regulation (EU) 2024/1689 (AI Act), Article 1 and recital 27. https://artificialintelligenceact.eu/article/1/ (accessed 2026-11-14)
  4. AI Act, recital 27 (seven non-binding ethical principles). https://artificialintelligenceact.eu/recital/27/ (accessed 2026-11-14)
  5. The Confiance.ai program reveals its first significant results, ActuIA, 6 October 2021 (13 structuring partners, 45 M EUR over 4 years). https://www.actuia.com/en/news/the-confiance-ai-program-reveals-its-first-significant-results-with-6-practical-use-cases/ (accessed 2026-11-14)
  6. Confiance.ai: a European roadmap for industrialising trustworthy AI, ActuIA, 10 April 2025 (programme ended in September 2024). https://www.actuia.com/actualite/confianceai-une-feuille-de-route-europeenne-pour-lindustrialisation-de-lia-de-confiance/ (accessed 2026-11-14)

Editorial methodCorrections log

← Back to glossary

Address copied