AI System

An AI system is, according to the AI Act, a machine-based system that can generate outputs such as predictions, recommendations or decisions. The definition determines whether your software is covered by the AI Act's requirements.

Back to Dictionary

The legal definition

The AI Act’s Article 3(1) defines an AI system as:

Definition: "A machine-based system that is designed to operate with varying levels of autonomy and 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."

This definition is intentionally broad. The EU has adapted it from the OECD’s definition of AI to ensure that it is technology-neutral and can accommodate future AI technologies.

The three key elements

The definition contains three decisive elements, all of which must be fulfilled:

1. Machine-based with autonomy: The system must be able to operate with a degree of independence. This means it can perform tasks without constant human control of every individual step. A calculator has no autonomy. A language model that formulates its own responses does.

2. Adaptiveness: The system must be able to adapt after deployment. This can occur through machine learning, where the model improves over time, or through inference, where the system adapts its output to new inputs. This characteristic distinguishes AI from traditional rule-based software.

3. Generates outputs: The system must produce predictions, content, recommendations or decisions. This covers everything from a chatbot generating text, to a system recommending credit scores, to a model predicting machine failures.

All three elements must be present. A spreadsheet with formulae is machine-based and generates outputs, but it lacks autonomy and adaptiveness. It is therefore not an AI system.

Examples of AI systems

Here are concrete examples of systems that typically fall within the definition:

Large language models (LLMs): ChatGPT, Claude, Gemini and similar. They are also general-purpose AI models with separate requirements.
Image recognition: Systems that identify objects, faces or defects in production lines.
Recommendation systems: Algorithms that recommend products, content or candidates for job interviews.
Predictive models: Systems that predict customer churn, credit risk or maintenance needs.
Autonomous vehicles: Self-driving cars and drones that navigate independently.
Robotic Process Automation with AI: RPA bots that use machine learning to handle unstructured data.

Once you have identified your AI systems, you should assess which risk categories they belong to. A high-risk AI system has significantly more requirements than one with minimal risk.

Borderline cases

Not all software with "AI" in its marketing material is an AI system in the legal sense. And conversely, software without AI branding may well be covered.

Typically NOT AI systems:

Simple rule-based systems (if/then logic without adaptiveness)
Traditional statistical software without autonomous decisions
Databases and search engines with simple matching
Basic automation such as e-mail autoresponders with fixed templates

Typically AI systems (even if not called AI):

Spam filters based on machine learning
"Smart" CRM systems that score leads automatically
Translation tools based on neural networks
Pricing algorithms that adapt to market data

If in doubt, we recommend mapping your systems. Examine the underlying technology and assess whether all three elements of the definition are fulfilled.

Frequently Asked Questions about AI System

What is an AI system according to the AI Act?

An AI system is a machine-based system designed to operate with varying levels of autonomy, and which after deployment can exhibit adaptiveness and generate outputs such as predictions, recommendations, decisions or content.

Is a simple rule-based system an AI system?

No, simple rule-based systems with fixed if/then rules are normally not AI systems within the meaning of the Regulation. The system must exhibit a degree of autonomy and adaptiveness to be covered.

Is a chatbot an AI system?

It depends on the technology. A chatbot based on large language models (LLMs) such as GPT is an AI system. A simple FAQ chatbot with fixed answers is typically not, because it lacks adaptiveness.

Why does the definition of AI system matter?

The definition determines whether your software falls under the AI Act’s requirements. If your system meets all three criteria (autonomy, adaptiveness, output generation), it is subject to the Regulation’s obligations depending on its risk category.