Exploring the Trustworthiness Challenges of Agentic AI at Data Week 2026

As artificial intelligence continues to evolve, so too do the challenges associated with ensuring it remains trustworthy, transparent, and accountable. At Data Week 2026 in Oslo, THEMIS Partner Professor Gregoris Mentzas, Director of the Information Management Unit (IMU) at ICCS and Professor at the National Technical University of Athens (NTUA), joined leading experts to discuss one of the most significant developments in the field: the emergence of Agentic AI systems.

THEMIS Partner, Professor Gregoris Mentzas, Presenting at the Closing Session of Data Week 2026

Organised by the Big Data Value Association (BDVA), Data Week is Europe's premier event for Big Data and Data-Driven AI research and innovation, bringing together researchers, industry leaders, policymakers, and innovators to explore the future of artificial intelligence and data technologies. Professor Mentzas participated in the conference's closing plenary panel, which focused on cutting-edge developments in AI and the implications of increasingly autonomous systems.

Data Week: From Generative AI to Agentic AI

A key message from Mentzas' presentation was that the transition from Generative AI to Agentic AI represents far more than a technological upgrade.

"The shift from Generative AI to Agentic AI is not incremental, it is structural."

While Generative AI systems are designed to produce content in response to prompts, Agentic AI systems are capable of perceiving their environment, planning actions, and operating autonomously over extended periods to achieve goals. This changes the nature of the governance challenge. Rather than focusing solely on the outputs generated by AI systems, organisations must increasingly consider the consequences of autonomous actions taken by AI agents.

New Risks Require New Approaches

Professor Mentzas highlighted several emerging risks associated with Agentic AI that existing governance frameworks were not originally designed to address. These include:

• Goal misalignment between human intentions and AI objectives

• Cascading failures across multi-agent systems

• Indirect prompt injection attacks

• Persistent memory poisoning

• Unpredictable interactions between autonomous agents

  
  

As AI systems become more autonomous, these challenges create new requirements for monitoring, oversight, and accountability.

Research Challenges Ahead

The discussion also explored the significant research challenges that must be addressed to ensure Agentic AI systems remain trustworthy and aligned with human values. Among the key priorities identified were:

• Runtime goal alignment

• Balancing autonomy and human control

• Dynamic auditing throughout system operation

• Responsibility attribution

• Behavioural benchmarks that go beyond traditional measures of accuracy

These challenges are closely aligned with the goals of our THEMIS 5.0 project, which is developing methods, tools, and approaches to help organisations assess and improve the trustworthiness of AI systems in real-world settings.

Advancing Trustworthy AI

The panel provided an fascinating opportunity to exchange perspectives with experts from across Europe and to discuss how research, industry, and policymakers can work together to address the next generation of AI governance challenges. Professor Mentzas thanked the organisers and fellow panellists for the engaging discussion, including Till Christopher Lech for the invitation, moderator Laure Le Bars, and panellists Kerstin Bach, Andrejs Vasiļjevs, and Signe Riemer-Sørensen. He ended on a call to action to the audience - 'Let's connect on advancing responsible and trustworthy Artificial Intelligence!'

Panellist at the Closing Plenary