Does the AI Act make my grid or asset AI high-risk?

Critical infrastructure. The AI Act takes the definition of Directive (EU) 2022/2557 on the resilience of critical entities (Article 3(62)): an asset, a facility, equipment, a network or a system, or a part of one, that is necessary for the provision of an essential service. Annex III point 2 covers only three areas of it: critical digital infrastructure, road traffic, and the supply of water, gas, heating or electricity.

Safety component. A component whose intended purpose is to prevent or mitigate risks to the health and safety of persons or property, or whose failure would endanger them (Article 3(14)). Systems used only for user assistance, performance optimisation, service efficiency, automation or quality control are not safety components, unless their failure would endanger health and safety (Article 6(1a) and 6(1b)).

Critical entity. An operator identified by a Member State under Directive (EU) 2022/2557; Member States had to identify them by 17 July 2026. According to the Commission's draft guidelines, point 2 applies only when the AI system is used by a critical entity (paragraph 190).

Official sources.

• Regulation (EU) 2024/1689, the AI Act, and Regulation (EU) 2026/1744, the Digital Omnibus on AI (EUR-Lex, in every EU language)
• Directive (EU) 2022/2557 on the resilience of critical entities
• Commission draft guidelines on the classification of high-risk AI systems, 19 May 2026
• European Commission, AI Act policy page
• AI Act Service Desk compliance checker, for the parts of the AI Act this tool does not assess

1. Describe the system. How it produces results (rules, statistics, a trained model, a language model), where the output goes, which decisions it supports, whether independent protection acts if it fails, and how it is presented in marketing and contracts.
2. Say who uses it. The sectors of the users and whether any of them is identified as a critical entity by a Member State. The check also asks about chat, generated text, the use of a general-purpose model and the dates of placing on the market.
3. Read the result. A live summary gives the likely classification. The report lists the reasons with their articles, the duties with the date from which each applies, read against the assessment date, and the next steps for the classification file. You can print it or save it as PDF.

Seetalabs keeps nothing: the answers stay in your browser.

The check covers one high-risk route: Annex III point 2, AI systems used as safety components in the management and operation of critical digital infrastructure, road traffic and the supply of water, gas, heating or electricity. It also flags, without assessing them, the prohibited practice of emotion recognition at work (Article 5(1)(f)), the biometric and employment areas of Annex III and the Annex I route for regulated products; for those, the result points to the Commission's compliance checker.

• Scope and definition. Article 2(1) and 2(8) of Regulation (EU) 2024/1689, with the roles of Article 3 (provider, deployer, importer, distributor, authorised representative); recital 12 and the Commission guidelines on the definition of an AI system, C(2025) 5053, paragraphs 46 to 51, for rules written by people and simple statistics.
• Safety component. Articles 3(12), 3(14), 6(1a) and 6(1b) as amended by Regulation (EU) 2026/1744; the draft guidelines on high-risk classification of 19 May 2026, paragraphs 180 to 186 and the electricity examples of section 3.2.7.
• Critical entities. Directive (EU) 2022/2557 and paragraphs 190 and 191 of the draft guidelines, which apply point 2 only to users identified as critical entities.
• Exemptions and transition. Article 6(3) and 6(4), paragraphs 71 and 108 of the draft guidelines, Article 111(2).
• Duties. Articles 4, 5(1)(f), 16, 22 to 27, 43(2), 47 to 49, 50, 53 and 72 to 73, with the dates of Article 113 as amended; guidelines C(2025) 7719 on general-purpose AI models.

Where the line is thin. The draft guidelines place anomaly detection that supports decisions on load, stability or shutdown inside point 2, and a precautionary warning to the control centre that does not act on itself outside. When the answers sit between the two, the result says so instead of choosing.

The draft guidelines are not final; the Commission expects the final version by the end of 2026. The result is an orientation, not legal advice.

The four examples built into the tool, assessed on 16 October 2026. They are fictitious systems, chosen to show where the line falls.

One answer is not enough to move the second example: it stays borderline without the alarm, and without the de-energising decision. With neither, so that the system only informs diagnosis and planning, it becomes "likely not high-risk".