What Counts as an AI System? The EU AI Act Definition and Commission Guidelines

Before an employer can ask what the EU AI Act requires of a tool, it has to ask a simpler question: is this software an "AI system" at all? The definition in Article 3 of Regulation (EU) 2024/1689 decides whether the Act's rules are in play, and the European Commission has published guidelines to help people apply it. This guide walks through the definition, the seven elements the Commission's guidelines identify, and the kinds of systems the guidelines place outside it. It is general information, not legal advice, and it is written for finance, HR and operations teams in Dublin and across Ireland. The application timetable for the wider Act is a separate matter, covered in our guide to the EU AI Act key dates for employers in Ireland.

Why the definition matters

The Commission's page announcing the guidelines (published 6 February 2025, last updated 8 April 2026) says that the AI Act's first rules started to apply on 2 February 2025. It lists them as the AI system definition, AI literacy and a very limited set of prohibited AI practices. The same page says the guidelines are meant to help providers and other relevant persons decide whether a software system is an AI system, so that the rules can be applied effectively.

That matters because everything else in the Act hangs off the definition. If a tool is not an AI system, the Act's AI-specific rules do not attach to it. If it is, the next questions are what role your organisation has and whether any further category applies. The guidelines themselves stress that meeting the definition does not trigger obligations on its own: only certain AI systems are subject to regulatory obligations and oversight under the Act, and most systems that qualify as AI systems will not be subject to any regulatory requirements under it. Dates for the other provisions are not repeated here; check the current position on the Commission's pages before relying on any date.

The definition in Article 3(1)

The Commission's guidelines (reference C(2025) 5053 final, dated 29 July 2025, hosted on the Commission's AI Act Service Desk) quote Article 3(1) in full. In summary, an "AI system" is a machine-based system that is designed to operate with varying levels of autonomy, that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations or decisions that can influence physical or virtual environments.

Recital 12 of the Regulation, which we read on EUR-Lex, explains the thinking. It says the notion should be clearly defined and based on key characteristics of AI systems, and it identifies the capability to infer as a key characteristic. It also says the definition should not cover systems that are based on rules defined solely by natural persons to automatically execute operations.

The seven elements in the Commission's guidelines

The guidelines break the definition into seven elements. In plain terms:

• Machine-based system: the system runs on hardware and software.
• Autonomy: the system is designed to operate with varying levels of autonomy, which Recital 12 describes as some degree of independence of actions from human involvement.
• Adaptiveness: the system may change its behaviour after deployment. The guidelines treat this element as optional, which tracks the words "may exhibit" in the definition.
• Objectives: the system pursues explicit or implicit goals.
• Inference: the system derives outputs from the inputs it receives, using AI techniques in the building phase.
• Outputs: predictions, content, recommendations or decisions.
• Interaction with the environment: the outputs can influence physical or virtual environments.

The inference element does most of the work in practice. A system that only follows instructions someone wrote out step by step is, on the guidelines' reading, in a different place from one that derives how to produce outputs from the data it receives.

What the guidelines place outside the definition

The guidelines include a section headed systems outside the scope of the definition. It covers the following, although for classical heuristics the text describes the method rather than stating an exclusion in terms:

• Systems for improving mathematical optimisation: systems that improve or approximate established optimisation methods, such as linear or logistic regression, which the guidelines say do not go beyond basic data processing.
• Basic data processing: systems that follow predefined, explicit instructions, with no learning, reasoning or modelling at any stage.
• Classical heuristics: experience-based problem-solving methods that apply predefined rules rather than data-driven learning. The guidelines describe these in the section on systems outside the definition.
• Simple prediction systems: systems whose performance can be achieved via a basic statistical learning rule, for example predicting an average or mean, are treated as outside the definition on the basis of their performance.

Two cautions apply. First, the guidelines say it is not possible to provide an exhaustive list of AI systems, and that the definition should not be applied mechanically: each system has to be assessed on its own characteristics. Second, the guidelines are not binding. They state that only the Court of Justice of the European Union can give an authoritative interpretation of the AI Act, and the Commission's page says the guidelines are designed to evolve and will be updated as necessary.

Why employers should care

According to the Commission's AI literacy questions and answers (page last updated 27 July 2026), Article 4 of the Act, as amended by the Digital Omnibus on AI, requires providers and deployers of AI systems to take measures to support the development of AI literacy of their staff and of other persons dealing with the operation and use of AI systems on their behalf. The page says no specific level of AI literacy is mandated. That duty only makes sense once you know which of your tools count as AI systems. Our guide to the Article 4 AI literacy duty covers what the Commission says about it.

The definition also feeds into governance. Someone in the business has to be able to say which tools are in scope and who is accountable for them, a theme in our piece on accountable AI use for managers.

A practical approach to scoping your tools

The following are our suggestions, not legal requirements:

• List what you use. Ask each team, in finance, HR, customer service and IT, which software makes predictions, rankings, recommendations, drafts or decisions. Include features inside products you already own, not just stand-alone AI tools.
• Walk each item through the seven elements. Write down, in a sentence each, why you think the tool does or does not infer outputs from inputs. If you cannot answer, that is a question for the supplier.
• Ask suppliers directly. Request a written statement of whether the supplier considers the product an AI system under the Act, and on what basis. Keep the answer with your records.
• Do not rely on the label. A product described as "smart" or "AI-powered" is not automatically an AI system, and a plain-looking feature might be. The guidelines call for a case-by-case assessment.
• Revisit when the guidelines change. The Commission says it will update them in light of practical experience, so note the version you relied on.

Treat the output as a working assumption, and take legal advice where a tool touches decisions about people, such as recruitment or monitoring.

Closing thought

The AI system definition is the gateway to the rest of the Act, and the Commission's guidelines are the most useful official reading aid available, even though they are not binding. For most employers the practical task is modest: know which of your tools might qualify, record your reasoning and keep an eye on the Commission's pages.

Learnsignal's offices are off Baggot Street in Dublin, and we offer in-person training from there. For current details, see our CPD pages.

Frequently asked questions

Is every piece of software with a statistical model an AI system?

No. The guidelines place certain systems outside the definition, including systems that improve established mathematical optimisation methods such as linear or logistic regression, and basic data processing that follows predefined instructions. They also say each system must be assessed case by case.

Are the Commission's guidelines legally binding?

No. The guidelines state that they are not binding and that only the Court of Justice of the European Union can give an authoritative interpretation of the AI Act. The Commission says they will be updated over time.

If a tool is an AI system, do all the Act's obligations apply?

No. According to the guidelines, only certain AI systems are subject to regulatory obligations and oversight, and most AI systems will not face regulatory requirements under the Act. Your role and the category of the system determine what applies, and the AI literacy duty in Article 4 is a separate point.

Does the definition require the system to learn after it is deployed?

No. The definition says a system "may exhibit adaptiveness after deployment", and the guidelines treat adaptiveness as an optional element.

Where should we check current dates?

Use the Commission's AI Act pages and EUR-Lex, and see our key dates guide. Check the current position before relying on any date, as the timetable can change.

In short: scope your tools against the definition, record why, and revisit when the Commission updates its guidance.