Self-Evolving Behaviour in Machinery: What It Really Means for AI Compliance

The EU Machinery Regulation has introduced a term that’s causing waves across the AI and industrial sectors: “self-evolving behaviour.” But what does it actually mean?

A recent Intertek blog breaks down the regulatory interpretation and it’s broader than many expect. It’s not just about AI systems that learn after deployment. Even pre-trained machine learning models that exhibit adaptive behaviour during operation may fall under this classification.

🔍 Key Takeaways:

• The regulation uses a capability-based approach, focusing on a system’s inherent adaptability—not just when learning occurs.
• AI systems with fixed parameters post-deployment may still be considered “self-evolving” if they were trained to respond flexibly to new inputs.
• This impacts a wide range of applications - from anomaly detection to predictive maintenance - where AI behaviour is shaped by prior learning.

⚠️ Why it matters:
This interpretation significantly expands the scope of compliance. Manufacturers must now prepare for enhanced safety evaluations, documentation, and validation, even for AI systems that don’t learn in real time.

As AI becomes more embedded in industrial machinery, understanding and addressing these regulatory expectations is critical to maintaining market access and ensuring safety.

🔗 Learn more about Intertek's AI Services, where our experts can assist you in complying with these new requirements:
AI²: https://www.intertek.com/ai/ 
Machinery Regulation: https://www.intertek.com/industrial-equipment/machinery-regulation/