How to decide if your system qualifies as an AI system?

Comparing definitions

To understand if a system or application qualifies as an AI system under Regulation (EU) 2024/1689 (the EU AI Act), we first need to unpack the definition.

EU AI Act definition

Article 3 of the EU AI Act defines an AI system as follows:

(1) ‘AI system’ means 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 closely related, though not identical to the OECD definition of an AI system.

OECD definition

The OECD reached consensus on the definition of an AI system in 2019 but the current definition, revised in 2023 is:

An AI system is a machine-based system 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. Different AI systems vary in their levels of autonomy and adaptiveness after deployment.

Image courtesy: OECD AI Principles overview. Retrieved February 2025.

Differences

The EU AI Act’s definition puts more emphasis on autonomy as a design principle and treats adaptiveness as an optional characteristic, aligning with its risk-based regulatory framework. Meanwhile, the OECD definition presents a more neutral and flexible view, making it more suitable for international policy discussions.

Interestingly, the words “designed to operate” were in the original OECD definition and removed in the 2023 revision.

European Commission guidelines

On February 6, 2025, the European Commission published guidelines on the AI system definition. The purpose of these guidelines is to help providers and others determine if a software system qualifies as an AI system under the AI act and is therefore subject to regulation. 

The guidelines are broken down into 7 sections, reflecting the characteristics we find in our definition:

1. Machine-based
2. Autonomy
3. Adaptiveness
4. AI system objectives
5. Inferencing how to generate outputs using AI techniques
6. Outputs that can influence physical or virtual environments
7. Interaction with the environment

Guiding questions to evaluate

For each characteristic, you can use the following questions to determine if a system qualifies as an AI system.

Machine-based system

• Is the system developed and executed on hardware and/or software?
• Does the system require computational processes to function?

Designed to operate with varying levels of autonomy

• Can the system function with minimal or no human intervention, or does it require manual operation at every step?
• Does it generate outputs that are not explicitly predetermined by humans?

May exhibit adaptiveness after deployment

• Can the system modify its behavior based on new data or interactions?
• Does the system update its models or decision-making rules without human reprogramming?
• Does the system evolve or improve over time in response to real-world conditions?

For explicit or implicit objectives

• Is there a specific goal or function that the system is designed to achieve?
• Are these objectives directly programmed, or do they emerge from system behavior?

Infers, from the input it receives, how to generate outputs

• Does the system process inputs to derive conclusions, predictions, or recommendations?
• Does it use techniques such as machine learning, logic-based reasoning, or statistical modeling?

Outputs such as predictions, content, recommendations, or decisions

• Are its outputs capable of influencing human decisions or automated processes?
• Can the outputs be classified as predictions, content generation, recommendations, or decisions that can influence human decisions or automated processes?

Can influence physical or virtual environments

• Do the system’s outputs impact the real world (e.g., robotics, medical devices) or digital spaces (e.g., search rankings, chatbots)?
• Does the system’s behavior have consequences beyond simple data processing?

Example: factory workforce planning

Imagine the following high-level description of a system and the context in which it is used.

When we review our guiding questions, we see that the example meets all 7 criteria and thus qualifies as an AI system under the AI act.

1. Machine-based system: The system is developed and executed on software, as it processes various inputs and integrates with other systems like holiday tracking. It meets this criterion.
2. Autonomy: The system functions with minimal human intervention, dynamically assigning shifts and adjusting schedules based on analyzed data. It meets this criterion.
3. Adaptiveness: The system analyzes historical patterns, trends, and external factors to optimize scheduling and predict absences. This indicates it modifies its behavior based on new data. It meets this criterion.
4. AI system objectives: The system is designed to achieve specific objectives, such as optimizing workforce scheduling and predicting absences. It meets this criterion.
5. Inferencing how to generate outputs using AI techniques: The system uses techniques like analyzing historical data, attendance trends, and external factors to infer optimal schedules and predict absences. It meets this criterion.
6. Outputs that can influence physical or virtual environments: The system generates shift assignments and schedule adjustments, which directly impact the workforce and factory operations. It meets this criterion.
7. Interaction with the environment: The system’s outputs influence the real-world operation of the factory by determining employee schedules and responding to dynamic factors. It meets this criterion.

Counter-example: simple timekeeping

Here’s an example of a system that doesn’t meet our requirements.

When we now review the main elements of our definition, we see that this example does not qualify as an AI system because it does not use specific AI techniques to infer its outputs.

1. Machine-based system: The system is developed and executed on software, as it processes clock-in and clock-out data to calculate hours worked. It meets this criterion.
2. Autonomy: The system functions with minimal human intervention, automatically calculating total hours worked. It meets this criterion.
3. Adaptiveness: There is no indication that the system modifies its behavior based on new data or interactions. It does not meet this criterion.
4. AI system objectives: The system is designed to achieve a specific objective—tracking hours worked and calculating pay. It meets this criterion.
5. Inferencing how to generate outputs using AI techniques: The system calculates hours worked based on predefined rules (clock-in and clock-out times). There is no evidence of inferencing or AI techniques being used. It does not meet this criterion.
6. Outputs that can influence physical or virtual environments: The system generates outputs (e.g., total hours worked and pay calculations) that influence payroll processes. It meets this criterion.
7. Interaction with the environment: The system’s outputs are limited to payroll calculations and do not have broader impacts on physical or virtual environments. It partially meets this criterion.

Ask our bot

Do you still struggle with deciding whether your system qualifies as an AI system?

Ask Governicus, our AI chatbot designed to assist with questions about AI governance, risk management, and compliance with the EU AI Act.

For example, use the following prompt:

Does this qualify as an AI system under the AI Act? 
"A company implements a simple employee timekeeping system. This software is used solely to track the hours worked by employees, ensuring that they are paid correctly for the time they’ve worked. The system allows employees to clock in and clock out using a digital interface, and it calculates the total hours worked each day based on a standard hourly rate."