High-risk use case · Annex III §2
EU AI Act — Critical infrastructure
AI as safety components in management and operation of critical digital, road, water, gas, heating, or electricity infrastructure.
Key obligations apply from
2 August 2026
Risk tier
high risk
Full AI Act applies by Aug 2027 for remaining Annex I scope
Scope notes: High-risk classification applies where AI failure could endanger life or physical integrity in critical infrastructure sectors.
First steps before 2 August 2026
- 1Map AI components in SCADA, grid, and transport safety systems
- 2Align AI risk management with existing NIS2 and sector safety regimes
- 3Establish post-market monitoring for safety-critical AI updates
Who supervises you?
Market surveillance authorities in each Member State enforce the AI Act. The European AI Office supports coordination, GPAI oversight, and cross-border cases.
AI Act implementation & AI OfficePriority mandates for you
3 instrumentsHigh-risk AI systems — requirements and conformity
Guidelines on the practical implementation of Chapter III requirements — risk management, data governance, transparency, human oversight, accuracy, and robustness (Art. 96(1)(a)).
Why it matters: Primary interpretive guide for Chapter III conformity once published — distinct from the Art. 6 classification consultation now open.
View on mapApplication of requirements to Annex III use cases
Guidelines on how Chapter III requirements apply to each category of high-risk AI listed in Annex III.
Why it matters: Maps abstract obligations to concrete use cases — essential for deployers in HR, credit, insurance, and biometrics.
View on mapPost-market monitoring plan template
Implementing act on the template for post-market monitoring plans required under Art. 72 for high-risk AI systems.
Why it matters: Part of your conformity assessment file — defines how you will collect and analyse performance data after deployment.
View on mapThis page is an independent orientation guide — not legal advice. National implementation and sector-specific rules may apply. Always verify scope against the official AI Act text and your supervisor.