TRUST-AI 2026 Brings AI Research into Focus with 29 New Publications
Researchers and practitioners working on trustworthy AI gathered in Bremen on 15–16 August for TRUST-AI 2026, the Second European Workshop on Trustworthy AI, held as part of IJCAI/ECAI 2026.
The workshop was co-organised by THEMIS partners, Gregoris Mentzas of the National Technical University of Athens, and Asbjørn Følstad of SINTEF, bringing together research on how trustworthy AI can move from high-level principles to approaches that can be assessed and applied in practice.
The two-day event received 45 paper submissions, with 29 papers accepted and 34 reviewers contributing to the selection process. The programme reflected the breadth of the field. Research covered fairness and bias, explainability, hallucination detection, robustness, risk assessment, regulation, generative AI and autonomous agents.
The accepted and published papers are available on the TRUST-AI 2026 Papers page.
Several contributions focused on the reliability of large language models, including methods for identifying hallucinations and improving confidence in AI-generated outputs. Others examined fairness in recruitment systems, multilingual model alignment and the use of trustworthy AI approaches in areas such as healthcare, education and industry.
Governance and regulation were also prominent themes. Presentations explored how organisations can assess AI risks in practice, including work related to the EU AI Act and the application of existing trustworthy AI assessment frameworks. A further session looked at robustness and security, highlighting the challenge of maintaining reliable AI performance when systems move from controlled environments into real-world use.
The rise of agentic AI added another dimension to the discussion. Research on autonomous and semi-autonomous systems raised questions around human oversight, accountability and the extent to which decision-making should be delegated to AI. Participants also took part in a group exercise on general-purpose AI in knowledge work, identifying key challenges and potential research priorities as these systems become more widely used in professional settings.
For THEMIS 5.0, the workshop provided an important opportunity to connect technical research with the human, organisational and regulatory dimensions of AI trustworthiness. Gregoris Mentzas also contributed to the research programme as co-author of work examining the assessment and mitigation of trustworthiness in educational AI.
Many project partners were on the programme committee spanning research, technology, policy and philosophy expertise: Dimitris Apostolou, Steve Taylor, Eleni Tsalapati, Giannis Stamatellos, Andrea Palumbo, as well as members from outside the THEMIS Consortium including Gabriel González Castañé and Henrik Junklewitz.
A recurring message across the event was that trustworthy AI cannot be treated as a fixed property of a system. It depends on context, use, users and impact, and must be assessed throughout the AI lifecycle.
TRUST-AI is expected to return for a third edition at ECAI 2027 in Athens, Greece, continuing the discussion as AI systems become more capable and more deeply embedded in real-world decision-making.
_____
TRUST-AI 2026 took place in Bremen, Germany, on 15–16 August 2026 as part of IJCAI/ECAI 2026. The workshop is supported by the European Union's Horizon Europe programme through the THEMIS 5.0 project (Grant Agreement No. 101121042).
Comments