Deploying Trustworthy AI Infrastructure: Lessons from Connected Health and Critical Services

Trustworthy AI is often discussed in terms of models, their accuracy, robustness, fairness or explainability. But real-world deployment makes trust an infrastructure challenge as well. Sensitive data may need to remain local, AI decisions must be observable and auditable, distributed components need to remain secure and resilient, and increasingly autonomous systems must operate within clear technical and organisational boundaries.

This session brings together three Horizon Europe projects to examine these challenges through the particular demands of healthcare and critical services. In connected health, CoGNETs explores how institutions can learn collaboratively from sensitive patient data while examining the consequences of faulty or compromised data sources. FRAME approaches healthcare from the perspective of agentic AI, investigating how local deployment, human oversight, tracing and operational evidence can support trustworthy clinical decision support.

ENACT provides a different critical-service perspective through a demanding live-event environment, where media, location and interaction data must remain synchronised while latency, demand and available computing resources continuously change. Its experience shows why trustworthy infrastructure also depends on continuous observability, controlled adaptation and the ability to verify the consequences of AI-driven infrastructure decisions.

By comparing lessons across these vertical environments, the session asks which requirements for trustworthy AI infrastructure are sector-specific and which are shared. The discussion will explore trade-offs around privacy, data governance, security, observability, autonomy and performance; the role of use-case partners in shaping technologies around real operational needs; and what is still required to move from research pilots towards dependable deployment.