Ethics Guidelines for Trustworthy AI and the Assessment List (ALTAI)

What it is

The Ethics Guidelines for Trustworthy AI were published on 8 April 2019 by the High-Level Expert Group on Artificial Intelligence (AI HLEG), an independent group of experts appointed by the European Commission. They set out what the group meant by trustworthy AI and how to achieve it.

The guidelines say trustworthy AI should be lawful, ethical and robust. They ground it in four ethical principles, which are respect for human autonomy, prevention of harm, fairness and explicability. From those they derive seven requirements that an AI system should meet throughout its lifecycle.

On 17 July 2020 the expert group presented the final Assessment List for Trustworthy AI, known as ALTAI. It translates the seven requirements into questions an organisation can work through for a given AI system. The Commission describes it as an accessible and dynamic checklist that guides developers and deployers of AI, and it is available both as a document and as a web-based tool.

Who it applies to

Both documents are voluntary. They create no legal obligation and have no enforcement. They are written for any organisation that designs, develops, deploys or uses AI, in the public or private sector, and they are used well beyond the EU.

In practice, organisations reach for ALTAI when they want a structured, recognised checklist for reviewing an AI system’s ethical and societal risks, often before they have a fuller management system in place. Public bodies have also used it as the basis for their own AI principles. Ireland’s Guidelines for the Responsible Use of AI in the Public Service use the same seven requirements as their principles.

What it requires

ALTAI is organised around the seven requirements. For each it asks a series of questions about how the system is designed, built and operated.

Human agency and oversight. Whether the system could affect people’s autonomy or decisions, whether users know they are dealing with AI, and what form of human oversight is in place, from human-in-the-loop to human-in-command.

Technical robustness and safety. Resilience to attack and misuse, fallback plans if the system fails, accuracy and how it is measured, and reliability and reproducibility.

Privacy and data governance. Protection of personal data, the quality and integrity of the data used, and who has access to it.

Transparency. Traceability of data and design decisions, explainability of outputs to the people affected, and communication about the system’s capabilities and limits.

Diversity, non-discrimination and fairness. Avoiding unfair bias in data and design, accessibility and universal design, and involving the people affected.

Societal and environmental well-being. Environmental impact, effects on work and skills, and wider effects on society and democracy.

Accountability. Auditability, how negative impacts are identified and minimised, how trade-offs are documented, and whether people have a route to redress.

Working through ALTAI produces a reflection on each requirement and a list of improvements. It is a self-assessment, not an audit, and it does not produce a rating.

Key dates

The guidelines were published on 8 April 2019, after a public consultation on a December 2018 draft. A piloting phase followed, and the final ALTAI was published on 17 July 2020. Neither document has been formally revised since. The European Commission’s page for ALTAI remains live and was last updated in November 2025.

How it relates to other frameworks

The seven requirements fed directly into the EU AI Act, which the expert group’s work preceded. The Act’s high-risk obligations on risk management, data governance, transparency, human oversight, and accuracy and robustness are the binding descendants of the ALTAI requirements. ALTAI is therefore a useful bridge for an organisation starting to prepare for the Act, but meeting ALTAI is not meeting the Act.

The guidelines appeared within weeks of the OECD AI Principles and cover much of the same ground. They are narrower than a management system standard such as ISO/IEC 42001, which sets out how to run governance across every AI system rather than how to review one. For a structured human rights assessment, the Council of Europe’s HUDERIA methodology goes further on scoring severity and planning mitigations.

Because the seven requirements are widely recognised, many organisations map their own AI principles or policy to them. A policy built that way is easier to explain to regulators and partners in the EU.