EU AI Act Annex III §2 — Critical Infrastructure (High-Risk)

Why this category is high-risk

Annex III §2 of Regulation (EU) 2024/1689 designates critical infrastructure AI systems as high-risk. Annex III §2 covers AI systems intended to be used as safety components in the management and operation of critical digital infrastructure, road traffic, and the supply of water, gas, heating and electricity. The risk profile is overwhelmingly safety-engineering rather than fundamental-rights — and the FRIA obligation under Article 27 is explicitly excluded for §2 systems.

What is covered

Examples in scope:

• AI safety components in electricity grid balancing and load-frequency control.
• AI safety components in gas-network pressure-control or leak-detection systems.
• AI safety components in water-supply pump scheduling or treatment-plant control.
• AI safety components in road-traffic management systems (signal coordination, incident detection on managed motorways).
• AI safety components in critical digital infrastructure operations (CDN routing decisions affecting essential services).

Out of scope of §2: AI used for billing, customer service, or commercial optimisation in these sectors. Those are not safety components and are not §2 high-risk (though they may be high-risk under other Annex III points).

Why §2 is different

§2 systems align with classical safety-critical engineering. The expected harmonised standards landscape leans on existing functional-safety frameworks — IEC 61508 series, sector-specific standards (IEC 61511 for process industries, IEC 62443 for industrial cybersecurity).

What providers must do — §2 specifics

• Functional-safety analysis (HAZOP, FMEA) integrated with the Article 9 risk register.
• Article 15 robustness testing against environmental perturbations and sensor-failure modes.
• Cybersecurity testing aligned with IEC 62443 where the system runs in industrial control environments.
• Coordination with sectoral regulators (energy regulators, transport authorities).

What deployers must do — §2 specifics

• Operator competence requirements under Article 26(2) align with sectoral certification regimes (e.g., transmission system operator personnel certification).
• No FRIA under Article 27 — explicitly excluded.
• Incident reporting under Article 79 sits alongside sectoral incident-reporting (e.g., NIS2 Directive obligations for digital infrastructure operators).

Obligations triggered

A system falling under §2 triggers the full Chapter III Section 2 + Section 3 stack:

• Article 9 — risk management system.
• Article 10 — data and data governance.
• Article 11 — technical documentation per Annex IV.
• Article 12 — automatic logging.
• Article 13 — transparency, instructions for use.
• Article 14 — human oversight.
• Article 15 — accuracy, robustness, cybersecurity.
• Article 16 — provider obligations.
• Article 17 — quality management system.
• Article 26 — deployer obligations.
• Article 43 — conformity assessment (Annex VI internal control for §2 except where stated otherwise).
• Article 47 — EU declaration of conformity.
• Article 48 — CE marking where applicable.
• Article 49 — EU Database registration.

Where the deployer is a public-sector body, a private entity providing public services, a credit-decision deployer (Annex III §5(b)), or a life/health insurance pricing deployer (Annex III §5(c)), the deployer also owes a FRIA under Article 27. Annex III §2 critical infrastructure is excluded from FRIA.

What providers must do

• Run the Article 9 risk management cycle continuously, with category-specific risks identified up-front (see "Worked risk rows" below in the related pages).
• Document the Article 10 data governance with disaggregated bias examination across sub-groups particular to this category.
• Build Article 14 human oversight measures into the system before placement.
• Validate Article 15 accuracy with disaggregated metrics.
• Complete the Annex IV technical file before placement.
• Sign the Article 47 declaration and register in the EU Database under Article 49.
• For non-EU providers: designate an authorised representative under Article 22 before placement.

What deployers must do

• Use the system in accordance with the Article 13 instructions for use.
• Assign Article 14 / Article 26(2) human oversight to competent, trained, authorised persons.
• Ensure deployer-controlled input data is relevant and representative.
• Monitor operation; inform the provider of incidents under Article 26(5).
• Where Article 27 applies, perform a FRIA.
• For public-sector deployers, register the deployment under Article 49 + Article 26(8).
• Inform workers of high-risk AI systems used at the workplace under Article 26(7) before put-into-service.

Inline crosswalk

• ISO/IEC 42001:2023 Annex A.5.2 — AI system impact assessment.
• ISO/IEC 42001:2023 Annex A.9.2 — Human oversight.
• NIST AI RMF MAP 1.1 — Intended purposes and context understood.
• NIST AI RMF MAP 5.1 — Likelihood and magnitude of impacts characterised.

Penalties

Article 99(4) — up to €15 million or 3% of worldwide annual turnover, whichever is higher.

---

Disclaimer. Reference; not legal advice. Verify with counsel. Reg text from Regulation (EU) 2024/1689.