Resource

AI in Railway Software: A CENELEC-Compliant Governance Framework

As railway organizations face pressure to reduce certification overhead and accelerate delivery, generative AI is entering the engineering conversation - but the regulatory stakes are high. EN 50716:2023 sets strict boundaries on where AI can and cannot operate in safety-critical software development, and misreading those boundaries carries real consequences for your Safety Case, your ISA relationship, and your SIL allocation. 

This white paper sets out a CENELEC-compliant framework for introducing generative AI on the development side of the V-model - where it can reduce traceability burden, enforce role independence, and improve review consistency - without crossing into the territory the standard explicitly closes. 

In this white paper you'll learn: 

  • Why EN 50716 Table A.3 prohibits AI/ML as a software architecture element at any SIL - and what that actually leaves open. 

  • How generative AI can defensibly assist requirements authoring, independent review, and traceability under EN 50716 tool qualification rules. 

  • How to enforce role independence, anti-hallucination protocols, and phase-gate criteria through tool configuration, not policy. 

  • What a compliant AI-assisted review process looks like from first draft to signed baseline. 

  • Where edge AI fits in rolling stock and infrastructure today - and where Annex C.3 still draws the line.