AI model or system? Identify your solution to comply with AI regulations

Since the adoption of the Artificial Intelligence Regulation (AIR), one of the most important and often overlooked aspects in the legal analysis of a technological solution is this: are we dealing with an AI model or an AI system?

This classification is not a minor detail. It determines the application of Regulation (EU) 2024/1689, the obligations of the various operators involved and, ultimately, the liability that a company may assume throughout the supply chain. In this article, we explain how to distinguish between the two concepts, AI model and AI system, and why this differentiation should always be the starting point in any legal analysis related to products based on artificial intelligence.

What do the Commission's Guidelines say about artificial intelligence systems?

Para facilitar esta tarea de diferenciación, la Comisión Europea ha publicado unas Directrices sobre la definición de sistema de inteligencia artificial, cuyo objetivo es ayudar a interpretar el concepto jurídico recogido en el artículo 3.1 del RIA.

Aunque no son vinculantes, estas Directrices descomponen la definición legal en siete elementos clave, explicando cómo debe entenderse cada uno y ofreciendo ejemplos prácticos. Según esta interpretación, una solución puede considerarse un sistema de IA si concurren (total o parcialmente) los siguientes componentes:

To facilitate this task of differentiation, the European Commission has published Guidelines on the definition of artificial intelligence systems, which aim to help interpret the legal concept set out in Article 3.1 of the AIA.

Although not binding, these Guidelines break down the legal definition into seven key elements, explaining how each should be understood and providing practical examples. According to this interpretation, a solution can be considered an AI system if the following components are present (in whole or in part):

1. Mechanical system (must be executed by machines, not biological processes).
2. Autonomy (operates with a certain degree of independence from humans).
3. Adaptability (can modify its behaviour once deployed).
4. Defined objectives (explicit or implicit in its design).
5. Inference capability (uses AI techniques to generate results from input data).
6. Generation of results with an impact on the environment (such as predictions, recommendations or decisions).
7. Interaction with the physical or virtual environment (its results have real, direct or indirect effects).

The Guidelines themselves clarify that they seek to offer a common interpretation that facilitates their practical application, without replacing or limiting the provisions of the legal text.

H2 Difference between AI model and AI system

What is an AI model?

In general terms, an AI model is a set of algorithms and mathematical structures trained with data to perform specific tasks, such as classifying information, making predictions or generating content. It is, so to speak, the component that provides ‘intelligence’ to a solution, but on its own it does not constitute a final product that is accessible or directly usable by a user.

For example, a model such as GPT-4 can generate text autonomously, but it does not have an interface, control mechanisms or business logic: it needs to be integrated into a broader solution to offer a functional service. It is, in essence, the technical basis, but not the finished product.

So what is an AI system?

An AI system, on the other hand, is a complete and operational solution. It is based on one or more AI models, but goes further: it includes elements such as user interface, operating rules, automated processes, monitoring mechanisms and updates, among others.

Its purpose is to fulfil a specific objective in a given environment and generate results that can influence (directly or indirectly) the behaviour of people, systems or processes. In this sense, it is a functional product, designed to interact with the environment and offer a useful result, not a simple technical tool.

Why is it so important to differentiate between the two?

The distinction between AI models and AI systems has specific legal implications within the framework of the AIA. Correctly identifying the nature of a product is essential in order to know:

• Which rules apply. The Regulation is primarily aimed at AI systems, except in the specific case of general-purpose models.
• What role the company plays. Only in the case of a system can it be clearly determined whether the organisation acts as a supplier, distributor, importer or deployment manager, with the corresponding legal obligations.
• What requirements must be met. From transparency and traceability to safe design or risk management, requirements vary depending on the type of product and the role of the operator.

Failure to correctly distinguish between the two categories can lead to undue obligations being assumed or, worse still, essential legal responsibilities being ignored.

Practical tips for distinguishing between model and system

What do we recommend?

Before deciding whether to sell, buy or integrate software that incorporates artificial intelligence, there is one question that needs to be clearly answered: are we dealing with a model or an AI system?

Regardless of the term used, what really matters is understanding how that solution is structured, what elements it comprises and how they interact with each other. This distinction is not merely technical: it is the first step in determining what obligations you will have as a company, what your role will be in the value chain and what risks you must anticipate and manage.

Understanding this structure (and how it fits into the European regulatory framework) is the first step in ensuring solid regulatory compliance, avoiding mistakes and building a legally sound technology strategy. Our lawyers who are experts in artificial intelligence can help you with this. Contact them here.

Data Protection and Digital Law Area