Policy brief “AI Ethics from Principles to Practice”

The AI Act establishes what organisations must achieve in AI ethics but not how. AIOLIA demonstrates that the way obligations like “ensure human oversight” or “provide explainability” translate into practice varies enormously by context, sector, and AI system type.

The European AI Office and national supervisory authorities should develop sector-specific, implementation-driven guidance, drawing on practitioner co-creation processes like AIOLIA’s, that shows organisations concretely how to operationalise each requirement, not merely what the requirement says.

AIOLIA’s bottom-up process reveals important gaps in ALTAI that limit its usefulness for practitioners. ALTAI does not adequately distinguish between professional and personal AI use contexts or account for the long-term and cumulative effects of AI on human cognition and behaviour, such as deskilling, dependency, and gradual manipulation.

It further lacks guidance on the context-specificity of ethics implementations across the AI lifecycle and does not adequately account for tensions when implementing competing ethics principles. The European Commission should provide accompanying guidance to ALTAI, treating AIOLIA’s D3.1 and D3.2 as direct inputs, to address these gaps.

The AI Office should consider developing ALTAI variants for specific high-risk AI systems under Annex III of the AI Act.

Small and medium-sized organisations often lack the capacity to build comprehensive ethics governance processes from scratch. While Article 57 of the AI Act requires member states to establish regulatory sandboxes, these are primarily framed around legal compliance testing.

Policymakers should expand the scope and resourcing of sandboxes, and leverage Digital Innovation Hubs, to offer structured support for ethics operationalisation, including access to multidisciplinary expertise, co-creation facilitation, and independent review mechanisms that are affordable to smaller actors.

The AI Act focuses primarily on the outputs and properties of AI systems, such as accuracy, explainability, and non-discrimination. AIOLIA demonstrates that the effective implementation of ethics depends strongly on the process through which ethics is operationalised: who is involved, how trade-offs are managed, how feedback is collected, and how findings are used to update systems and procedures.

Policymakers should introduce process-level requirements for high-risk AI — mandating co-creation with affected communities, multidisciplinary governance committees, feedback loops, and documented ethics decision-making — alongside existing output requirements.

AIOLIA’s co-creation process, pairing academic ethics and AI expertise with industrial partners through structured operationalisation cycles, produced 175 concrete measures that neither party could have developed alone. AIOLIA has condensed them into a 6-page Measures Portfolio. This Portfolio generates ethics guidance that is simultaneously principled and practically grounded.

Policymakers should recognise co-creation partnerships as a legitimate and fundable approach to AI ethics implementation, and create EU-level infrastructure, including funding, coordination, and methodology toolkits, to make this model accessible beyond research projects.

AIOLIA identifies a critical knowledge gap: long-term effects of AI systems on human cognitive capacities, decision-making patterns, social skills, and wellbeing are poorly understood. Without this evidence base, organisations and policymakers are making precautionary judgements without solid empirical basis.

The EU should fund structured longitudinal research programmes, enabling independent researchers to access operational AI systems and usage data, with robust data protection safeguards, to build the evidence base needed for informed, proportionate regulation.