Luca BertonManufacturing AI: When Safety Components Meet the AI Act
Manufacturing AI sits at a unique intersection of the EU AI Act and the Machinery Regulation (2023/1230). AI used as a safety component in machinery — from robotic arm collision avoidance to process control systems — is automatically classified as high-risk. But even non-safety AI like quality inspection and predictive maintenance may fall under specific obligations.
Risk Classification for Manufacturing AI
High-Risk: Safety Components (Annex I, Section A)
- Robotic safety systems — Collision detection, human proximity sensing, emergency stop AI
- Process control AI — Chemical process safety, pressure vessel monitoring, temperature control
- Autonomous vehicles — AGVs and AMRs in factory environments
- Structural monitoring — AI assessing structural integrity of equipment, bridges, buildings
Not High-Risk but Regulated
- Visual quality inspection — Defect detection on production lines (may still need transparency if affecting workers)
- Predictive maintenance — Equipment failure prediction (not safety-critical in most cases)
- Supply chain optimisation — Demand forecasting, inventory AI
- Energy optimisation — Factory energy management using AI
Machinery Regulation + AI Act Dual Compliance
The new Machinery Regulation (2023/1230), effective January 2027, explicitly addresses AI in machinery:
- Art. 5: AI-enabled safety functions must undergo third-party conformity assessment
- Annex III: Essential health and safety requirements now include software and AI behaviour
- Digital instructions: AI-powered machines can provide digital-only user instructions
- Substantial modification: Updating AI models in deployed machinery may trigger re-conformity assessment
Kubernetes Recipes
A practical guide for container orchestration and deployment by Grzegorz Stencel & Luca Berton (Apress).
Watch on Skillshare →Infrastructure for Manufacturing AI Compliance
- Edge deployment — Manufacturing AI often runs on-premises or at the edge. Compliance infrastructure must work in air-gapped environments.
- OT/IT convergence — AI bridging operational technology and IT networks needs specialised security architecture
- Real-time inference — Quality inspection AI on 100+ unit/minute production lines needs deterministic latency
- Model lifecycle on equipment — Updating AI on deployed machinery requires change management that satisfies both Machinery Regulation and AI Act
- Safety integrity levels — AI safety components may need to satisfy SIL (Safety Integrity Level) requirements from IEC 61508
Practical Steps for Manufacturers
- Classify all AI systems — Map each AI application to EU AI Act risk categories, paying special attention to safety components
- Audit existing CE marking — Machinery with AI may need updated conformity assessment under new Machinery Regulation
- Implement model governance — Version control, change management, and rollback for AI on production equipment
- Edge monitoring — Deploy compliance monitoring that works in factory environments (latency, connectivity constraints)
- Document, document, document — Technical documentation requirements are extensive for both regulations
Microsoft SQL Server Performance Tuning
Performance tuning essentials for SQL Server. In collaboration with Starweaver.
Start on Coursera →Related Solution
Navigating AI adoption in a regulated environment? Our readiness assessment maps infrastructure, governance, and compliance gaps in 3-4 weeks.
Explore AI Readiness for Regulated Enterprises →