Cyber, data & AI

AI system (artificial intelligence)

An AI system is a machine-based system that is designed to operate with varying levels of autonomy, may exhibit adaptiveness after deployment and, 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. This is the definition in Article 3(1) of the EU AI Act. That capability to infer distinguishes it from software that only executes rules laid down by people.

Source: Regulation (EU) 2024/1689 (AI Act) Article 3(1), (3) and (4) and recitals 12, 97 and 99; OECD AI Principles (2019, updated May 2024)

The definition sets the scope of the EU AI Act (Regulation (EU) 2024/1689). Recital 12 calls the capability to infer the key characteristic: the system derives models or algorithms from inputs or data, through machine learning or through logic- and knowledge-based approaches. Systems that only execute rules defined solely by natural persons fall outside it. Autonomy means a degree of independence from human involvement; adaptiveness means that the system can change while in use through self-learning. The Act seeks alignment with the work of international organisations, and the definition in the OECD AI Principles (2019, updated in May 2024) is almost identical.

For the management board and the supervisory board, the definition is the starting point of the AI inventory. Only once it is clear which applications are AI systems can the organisation determine its role for each system, as provider or as deployer (Article 3(3) and (4)), and establish the risk category. Purchased AI and AI embedded in other software belong in that inventory. In the financial sector, systems used to evaluate the creditworthiness or establish the credit score of natural persons, and systems for risk assessment and pricing in life and health insurance, are high-risk (Annex III, point 5); the obligations for them apply from 2 December 2027.

An AI system is not the same as an AI model. According to recital 97, a model is an essential component of an AI system, but it only becomes a system once other components are added, such as a user interface. Nor is every calculation model covered by a model risk policy an AI system: a model without the capability to infer falls outside the definition. Anyone assessing whether an application falls under the Act should document that judgement for each system.

Last updated: · Editor: ONE Risk Advisory

Frequently asked questions

Is all software that uses algorithms an AI system?

No. According to recital 12 of the EU AI Act, systems that only execute rules defined solely by people fall outside the definition. What matters is whether the system itself infers from its input how to generate output, through machine learning or through logic- and knowledge-based techniques. A decision tree drawn up by people, or a spreadsheet with fixed formulas, is therefore not an AI system.

What is the difference between an AI system and a general-purpose AI model?

A general-purpose AI model can perform a wide range of distinct tasks and can be integrated into many systems; large generative models are the typical example (recital 99). The model is a component and becomes an AI system with additions such as a user interface. The AI Act sets separate obligations for both; those for providers of general-purpose AI models have applied since 2 August 2025.

Does the AI Act apply if an organisation only buys and uses AI?

Yes. Anyone using an AI system under its own authority is a deployer (Article 3(4)) and has its own obligations, such as taking measures to support the AI literacy of staff (Article 4) and, for a high-risk system (from 2 December 2027 for Annex III), using it in line with the instructions for use and assigning human oversight to competent people (Article 26). Use in a personal, non-professional activity falls outside the definition of deployer.

← All terms