Designing Embedded Systems with AI inside - Thales UK
Overview
In this session from the CMO & CAIO Inspired Summit at Syon Park, London, in October 2026, Anthony Collins of Thales UK draws a clear line between publicly available AI, where the cost of failure might be a rewritten prompt, and mission-critical AI in areas such as aerospace, defence and critical national infrastructure, where failure can have serious consequences.
Using the analogy of a light switch that might turn on the wrong bulb, Anthony explains why AI moves systems from deterministic to probabilistic behaviour, and how Thales designs for that. He walks through a systems engineering approach that puts outcomes and requirements before technology choices, treats data as a component in its own right and wraps AI in guard rails, using an autonomous search and rescue drone as a worked example.
Key takeaways
- Define the system outcomes, requirements and explanation of what the AI does before choosing any technology, and verify and validate against them throughout.
- AI is probabilistic, so the risk of an unexpected output has to be quantified, explained and managed down to an acceptable level.
- Treat data as a system component with its own life cycle: training and validation data must genuinely represent the live data the system will see.
- Place guard rails around the AI layer, sanitising inputs and constraining outputs, and match the level of autonomy (human in the loop, human on the loop or full autonomy) to the cost of failure.
- Red teaming by people outside the design team helps uncover edge cases, such as a drone trained in Welsh rain meeting Scottish snow.
Watch the full session on YouTube: Designing Embedded Systems with AI inside


