What matters in AI.

Subscribe

Learn / AI governance

Definition · AI governance

Critical infrastructure

Critical infrastructure is the set of systems and assets that a country cannot lose. Their loss causes large damage to its security, economy, or public health and safety. Examples are electricity grids, water supply, road traffic and communications. In the EU AI Act, AI failure in it can put life and health at risk at large scale.

Last reviewed

Key points

  • In US law, critical infrastructure is systems and assets whose loss has "a debilitating impact" on security, the economy, or public health and safety.
  • The US has 16 sectors (CISA). The EU CER Directive (EU) 2022/2557 has 11 sectors.
  • Under the EU AI Act, AI that is a safety component in this infrastructure is high-risk (Annex III, point 2).
  • The Annex III rules apply from 2 December 2027 (European Commission, 27 July 2026).
  • In the US, the DHS framework on AI in this infrastructure is voluntary (November 2024). CISA and partners gave joint principles in December 2025.

How the law sets critical infrastructure

A law gives the test. Then a list gives the sectors.

United States European Union
Law 42 U.S.C. 5195c(e) CER Directive (EU) 2022/2557
Test “a debilitating impact” entities “critical or vital for society and the economy”
Sectors 16 (CISA) 11 (CER Annex)

The EU AI Act uses the CER definition (Article 3(62)). It uses the term in two places:

  1. High-risk list. Annex III, point 2 covers AI as a safety component. The areas are digital infrastructure, road traffic, and the supply of water, gas, heating or electricity.
  2. Serious incident. Article 3(49)(b): “a serious and irreversible disruption” of critical infrastructure.

Why it matters for AI security

An error in a chatbot gives a bad answer. An error in AI that controls a valve can put persons at risk. Recital 55 of the AI Act says a failure can put “the life and health of persons at large scale” at risk.

DHS names three types of AI risk in critical infrastructure (14 November 2024):

  • attacks that use AI
  • attacks on AI systems
  • errors in design and implementation

In practice

A high-risk system needs a conformity assessment and human oversight, with other duties. EU AI Act Article 6 gives the rules for the Annex III list. The AI Omnibus started on 27 July 2026, and the Annex III rules apply from 2 December 2027 (European Commission).

The 11 EU sectors of the CER Directive:

  • energy, transport, digital infrastructure
  • banking, financial market infrastructure
  • health, drinking water, waste water
  • public administration, space, food production and distribution

The US National Security Memorandum of 30 April 2024 replaces PPD-21 as the US policy. EU Member States had to identify their critical entities by 17 July 2026.

The guidance for operators

Document Who Date Status
DHS framework on AI in critical infrastructure DHS, AI Safety and Security Board 14 Nov 2024 Voluntary framework
Principles on AI in Operational Technology CISA, ACSC, agencies of 7 countries 3 Dec 2025 Joint principles
AI Act, Annex III point 2 European Union 2 Dec 2027 EU law

The DHS framework gives roles to five groups:

  • cloud and compute providers
  • AI developers
  • infrastructure owners and operators
  • civil society
  • the public sector

The joint CISA guidance covers machine learning, LLM-based AI and AI agents in operational technology (OT). Its four principles:

  1. Understand AI.
  2. Consider AI use in the OT domain.
  3. Establish AI governance and assurance frameworks.
  4. Embed safety and security practices into AI and AI-enabled OT systems.

Its authors also urge operators to “continuously monitor, validate, and refine” their models. For a method to manage these risks, see AI risk management.

Where definitions disagree

The US and EU lists are not the same. Some examples (CISA list and CER list):

  • The CISA list has Chemical, Dams, Critical Manufacturing and Defense Industrial Base sectors. The CER list has no sectors with these names.
  • The CER list has a space sector. The CISA list has no space sector.
  • Some areas have different names in the two lists: Government Services and Facilities (US), public administration (EU).

Annex III, point 2 of the AI Act is smaller than the CER list. It names only digital infrastructure, road traffic, and the supply of water, gas, heating and electricity. Banking and health are in the CER list, but not in that point.

Questions and answers

Is every AI system used by a power or water company high-risk under the EU AI Act?

No. Annex III, point 2 covers AI used as a safety component in the management and operation of the infrastructure. Recital 55 says a safety component protects the physical integrity of the infrastructure or the health and safety of persons. An AI that writes invoices for a water company is not one. An AI used solely for cybersecurity is also not a safety component.

Does the US have binding AI rules for critical infrastructure?

Not at the federal level, in the sources on this page. The DHS framework of November 2024 calls itself voluntary. The joint CISA guidance of December 2025 gives principles that its authors urge operators to use. Sector rules that are not specific to AI can still apply.

Is critical infrastructure the same in the US and the EU?

The idea is the same, the lists are not. The US has 16 sectors, which include chemical, dams and critical manufacturing. The EU CER Directive has 11 sectors, which include space. The CISA list has no space sector.

Sources

  1. 42 U.S. Code § 5195c, Critical infrastructures protectionLegal Information Institute, Cornell Law School
  2. Critical Infrastructure SectorsCISA
  3. Critical Infrastructure Security and ResilienceCISA
  4. Critical infrastructure resilienceEuropean Commission, Directorate-General for Migration and Home Affairs
  5. EU AI Act, Article 3Future of Life Institute (text of Regulation (EU) 2024/1689)
  6. EU AI Act, Annex IIIFuture of Life Institute (text of Regulation (EU) 2024/1689)
  7. EU AI Act, Recital 55Future of Life Institute (text of Regulation (EU) 2024/1689)
  8. AI Omnibus enters into forceEuropean Commission, 27 Jul 2026
  9. Groundbreaking Framework for the Safe and Secure Deployment of AI in Critical InfrastructureU.S. Department of Homeland Security, 14 Nov 2024
  10. Principles for the Secure Integration of Artificial Intelligence in Operational TechnologyCISA, 3 Dec 2025
  11. Principles for the secure integration of AI in OTNCSC New Zealand

In the news