Title:

On Trustworthy Autonomy of Cyber-Physical Systems

Keynote speaker:

Dr Yannick Fourastier, Technology Group EMEA, Aerospace& Defence, Cyient, United Kingdom

Dr Yannick Fourastier, based in United Kingdom, is currently a Senior Director Technology Group at Cyient. Yannick Fourastier brings experience from previous roles at Egis, Thales and CodEUrope.

Abstract:

Trust in artificial intelligence (AI) has become a central concern for dependable and autonomous systems engineering and is now a measurable property of engineered systems. Modern AI, whether predictive, generative, or decision-making, increasingly governs safety-critical cyber-physical systems (CPS) such as robotics, utilities, transportation, and defense platforms. As these systems gain autonomy, unsafe behaviour can have tangible consequences, and their decisions must therefore be demonstrably safe, predictable, and accountable. Yet current approaches to AI ethics, robustness, and verification remain fragmented and often fail to produce auditable assurance that extends across contexts and over time.

This talk argues that trustworthy autonomy must rest on a dual foundation: philosophical coherence, which defines what it means for a system to merit trust, and technical assurance, which demonstrates through evidence why it can be trusted. Despite significant advances in verification, robustness, and assurance, there is still no unified method for producing auditable, evidence-based trust claims that remain valid under different operational conditions. The talk introduces a method for constructing measurable trust claims, defined as verifiable statements linking hazards, properties, and evidence throughout the AI lifecycle.

It proposes an epistemological and methodological synthesis that defines AI trustworthiness as the capability to generate verifiable claims of safety, reliability, and ethical compliance grounded in measurable evidence. By integrating epistemic justification with runtime verification and continuous evaluation, the paper outlines how AI systems can evolve toward self-maintaining dependability: systems able to monitor, explain, and justify their own trustworthiness. The analysis connects historical paradigms with contemporary AI safety research and concludes with an agenda for building a closed, self-sustaining ecosystem of trustworthy autonomy in cyber-physical applications.

Keywords: AI trustworthiness, Dependable autonomous systems, Trust claims and evidence, Runtime assurance, AI safety and verification, Epistemological foundations of trust, Technical assurance, Self-maintaining dependability.