Trustworthy AI tools overlook explainability and security, new study finds

Researchers from the National Technical University of Athens and the National Centre for Scientific Research 'Demokritos' published a critical analysis of trustworthy AI (TAI) tools and mark frameworks on arXiv. Leveraging a comprehensive dataset from the OECD, the team empirically mapped and compared over 100 tools and certifications. Their analysis reveals significant asymmetries: fairness, transparency, and robustness dominate the landscape, while explainability, digital security, and environmental sustainability receive disproportionately little attention. Furthermore, most tools concentrate on post-development validation and monitoring, leaving early design phases and data collection largely unguided. Educational initiatives and policy engagement are also underdeveloped, suggesting TAI efforts are currently skewed toward technical and procedural measures within industry contexts.

The paper argues that this implementation chasm between high-level ethical principles and practical deployment poses real risks—from biased systems to insecure deployments. The authors recommend expanding ethical objectives beyond a narrow set, embedding ethics across the full AI lifecycle (design, data, development, deployment, monitoring), and fostering broader multi-stakeholder participation including civil society and regulators. They provide actionable recommendations for more holistic, inclusive, and enforceable AI governance. As AI systems increasingly impact society, this study offers both a diagnosis of current gaps and a roadmap for closing them, emphasizing that trustworthy AI requires more than technical fixes—it demands systemic change in how ethics are operationalized.