Is all software an AI system under the EU AI Act?

No. Systems that produce outputs purely by executing rules defined by natural persons are outside the definition. The distinguishing feature is inference — deriving how to generate an output from data rather than following an explicitly specified rule.

Is a general-purpose AI model the same as an AI system?

No, and the Act treats them separately. A GPAI model is a model; an AI system is the system that gets placed on the market or put into service. A model becomes part of an AI system once it is integrated into one, and the two carry different obligation sets.

Who decides whether our product is an AI system?

You do, in the first instance, and you are expected to be able to defend it. National market-surveillance authorities can disagree and reclassify, so the safe practice is to document the assessment — the objectives, the inference step, the autonomy — at the point the product is designed.

EU AI Act glossary · Core concepts· Art. 3(1)

AI system

An AI system is a machine-based system that, for explicit or implicit objectives, infers from its input how to generate outputs such as predictions, content, recommendations or decisions that can influence physical or virtual environments.

The definition in Article 3(1) is deliberately broad and technology-neutral, aligned with the OECD definition. Its load-bearing word is inference — the capacity to derive outputs from inputs in a way that goes beyond a rule a human wrote out in full. That is what separates an AI system from ordinary deterministic software, and it is why the Commission's guidance on the definition focuses on whether a system learns or reasons rather than on which architecture it uses.

Where AI system sits in the Regulation

Art. 3(1)
Recital 12

Who it binds

AI system: why it matters

This is the on/off switch for the entire regulation. If your software is not an AI system (and is not a general-purpose AI model), no risk tier, no conformity assessment and no transparency duty can attach to it. That makes the definition the single most consequential scoping decision a compliance team makes, and it is a decision worth writing down with reasons — a regulator asking why a system was excluded will want the analysis, not a conclusion.

Worked example

AI system in practice

A pricing engine that applies a published rate card with fixed if/then rules is not an AI system: every output is traceable to a rule a human authored. Swap in a gradient-boosted model that learns price elasticity from historical transactions and it becomes one, because the mapping from input to output was inferred rather than specified. Nothing about the user interface changed — the classification turned entirely on how the output is produced.

Correction

AI system: the mistake people make

Assuming "AI system" means machine learning or generative AI only. The definition is deliberately architecture-agnostic and can reach knowledge-based, logic- and search-based approaches; conversely, calling a product "AI-powered" in marketing copy does not make it one in law. Classify on how outputs are actually generated, not on the label.

Questions

AI system: common questions

Is all software an AI system under the EU AI Act?: No. Systems that produce outputs purely by executing rules defined by natural persons are outside the definition. The distinguishing feature is inference — deriving how to generate an output from data rather than following an explicitly specified rule.
Is a general-purpose AI model the same as an AI system?: No, and the Act treats them separately. A GPAI model is a model; an AI system is the system that gets placed on the market or put into service. A model becomes part of an AI system once it is integrated into one, and the two carry different obligation sets.
Who decides whether our product is an AI system?: You do, in the first instance, and you are expected to be able to defend it. National market-surveillance authorities can disagree and reclassify, so the safe practice is to document the assessment — the objectives, the inference step, the autonomy — at the point the product is designed.

Keep reading

AI system: terms that come up alongside it

General-purpose AI (GPAI) modelArt. 3(63)

The foundation-model category with its own obligation set under Articles 53 to 55.

High-risk AI systemAnnex III / Art. 6

The tier where the Act's full compliance machinery applies, from Annex III and regulated products.

EU AI ActReg. (EU) 2024/1689

The EU's risk-tiered AI regulation, phased in from 2025 to 2028.

See how AI system applies to your systems

Run the free 2-minute EU AI Act checker — no sign-up.

Run the free check