Policy Brief: Building Trustworthy AI with Practical Recommendations from THEMIS 5.0
As artificial intelligence (AI) becomes more and more embedded in organisations, public services and high-impact settings, questions of trust are becoming just as important as questions of performance. An AI system may be accurate or technically robust, but that does not automatically make it trustworthy.
The THEMIS 5.0 innovation project has explored this challenge from a broader socio-technical perspective, looking at how people, organisations, governance structures, legal requirements and ethical considerations all shape the way AI systems are developed, deployed and experienced.
To make these findings accessible to different audiences, project partners have leveraged evidence from the project to produced two complementary resources:
Full Policy Brief (Deliverable D6.3) – Socio-Political, Legal & Ethical Recommendations a deep dive for those who want to explore the evidence, methodology and lessons behind THEMIS' socio-technical recommendations
Short Policy Brief - Building Trustwortrhy AI in Complex Socio-Technical Systems for readers just looking for the key messages and recommendations at a glance.
Why do we need a broader approach to trustworthy AI?
One of the central messages from THEMIS 5.0 is that trustworthiness cannot be reduced to technical performance alone. The Policy Brief highlights that concepts such as fairness, accuracy and robustness need to be understood in relation to the particular sector, use case and deployment environment. It also stresses the importance of governance, organisational readiness, user skills, operational practices and ethical and legal oversight.
This matters because AI systems do not operate in isolation. They function within wider environments shaped by people, organisations, regulations, operational processes and governance structures. As a result, the impact of an AI system depends not only on how the technology has been designed, but also on how it is implemented, used and managed.
The Brief turns the lessons and best practices from across the THEMIS 5.0 project into 22 recommendations for the development and deployment of trustworthy AI. These recommendations focus particularly on defining trustworthiness in socio-technical environments, incorporating user values and preferences, and improving the assessment and mitigation of risks.
What does THEMIS 5.0 recommend?
A recurring theme is the importance of participation. Trustworthy AI should not be developed only by technical teams and then presented to users at the end of the process. The Policy Brief recommends involving users, employees and other affected stakeholders throughout development, deployment and use. It encourages developers to use co-creation insights early, involve a broad range of stakeholders and treat participatory technology assessment as an ongoing process rather than a one-off exercise.
The project also emphasises that trustworthiness is context-specific. Standard metrics alone may not capture how fairness, robustness or other characteristics are understood by the people and organisations affected by an AI system. The recommendations therefore call for social and legal understandings of trustworthiness to be considered alongside technical ones, and for fairness to be defined in relation to a particular use case rather than treated as a universal score.
Another important area is ethical and legal assessment. THEMIS 5.0 recommends continuous assessment of ethical, legal and societal risks throughout the AI lifecycle, rather than treating these questions simply as a final compliance exercise.
Who should read the AI Policy Brief?
The findings are relevant to a wide range of people involved in AI.
For policymakers and research funders, the recommendations offer ideas for supporting participatory and multidisciplinary approaches through funding, programmes and guidance.
For developers and technical professionals, they highlight why stakeholder perspectives, organisational context and social and legal considerations need to be incorporated alongside technical measures.
For organisations adopting or deploying AI, the work underlines the importance of listening to employees, investing in user skills and governance, and continuing to manage trustworthiness after deployment.
For legal professionals, ethicists, social scientists and AI governance specialists, the publications show why different forms of expertise need to be brought together when assessing the opportunities and risks associated with AI.
A shared challenge
The central lesson is that trustworthy AI is not achieved through technology alone. It requires collaboration between people, organisations and institutions; meaningful stakeholder participation; an understanding of the context in which AI is used; and ongoing attention to governance, ethical considerations and legal risk. This is a move beyond technical compliance towards making trustworthiness a core policy, governance and organisational objective.
Whether you are shaping policy, developing technology, funding research, assessing AI systems or introducing AI into an organisation, the Policy Brief in whichever form you choose, offers practical insights into what building trustworthy AI can look like in practice.
Read the Short Policy Brief → Policy Brief - Building Trustworthy AI in Complex Socio-Technical Systems | Zenod
Read the Full Deliverable D6.3 → Full Policy Brief: Socio-Political, Legal & Ethical Recommendations (for building trustworthy AI) | Zenodo
Comments