18/04/2026
Spain is moving towards its own model of artificial intelligence governance: between European harmonisation and domestic challenges
By Iván Martínez, CEO of Intedya
The process of regulating artificial intelligence in Europe is entering a decisive phase, and Spain is already positioning itself for the next stage of regulation: national adaptation. Opinion 3/2026 of the Economic and Social Council (CES) analyses the Draft Bill on the proper use and governance of artificial intelligence, a key piece of legislation which does not introduce a comprehensive autonomous regulatory framework, but rather acts as an internal mechanism of Regulation (EU) 2024/1689, the so-called EU AI Act. This regulatory development is no small matter; in practice, it involves building the institutional, enforcement and operational framework that will enable the European risk-based model for artificial intelligence to be applied in Spain.
A national framework serving a directly applicable European regulation
The Draft Bill responds to a clear obligation: Member States must put in place the necessary national elements to make the European legislation operational. In this case, the central objective is twofold: on the one hand, to establish the system for the governance and supervision of artificial intelligence in Spain and, on the other, to define the penalty regime applicable to breaches of the European Regulation within national territory. It is not, therefore, a substantive law on artificial intelligence in the broad sense, but rather a ‘legal infrastructure’ framework: it designates authorities, sets out coordination mechanisms, establishes procedures and specifies how infringements will be penalised. This approach is consistent with the nature of the AI Act, which is directly applicable but requires a national layer for its effective implementation.
Governance: a complex and distributed institutional architecture
One of the most significant elements of the Draft Bill is the definition of the system of competent authorities; far from opting for a purely centralised model, Spain has established a hybrid framework. The Spanish Artificial Intelligence Supervisory Agency (AESIA) emerges as a central player, acting as a supervisory authority across multiple areas and as a single point of contact; however, it does not have a monopoly on supervision; other sectoral authorities, such as the Bank of Spain, the CNMV or the Spanish Data Protection Agency, retain powers over high-risk systems within their respective fields. This design reflects a classic tension in technology regulation: the need for sector-specific expertise versus regulatory consistency. The CES, in fact, warns of the risk of fragmentation and emphasises the importance of effective mechanisms for inter-institutional coordination. To mitigate this risk, the draft bill introduces a joint commission for collaboration between authorities, aimed at the exchange of information and best practice. However, the report itself suggests that this provision could prove insufficient unless coordination with other institutional actors – including those in the economic, labour and social spheres – is strengthened.
The European risk model: the structural backbone of the system
Although the draft bill does not redefine the regulatory model, it does put it into practice. The system is based on the risk-based approach of the European Regulation, which distinguishes between prohibited practices, high-risk systems and other systems subject to transparency obligations; this approach entails a graduated regulatory response: the greater the potential risk to fundamental rights, the greater the burden of obligations. The CES emphasises that this model seeks to balance two objectives that are not always compatible: the protection of fundamental rights and the promotion of technological innovation. In this regard, the national legislation does not introduce any significant deviations from the European framework, but rather implements it faithfully.
Penalty regime: alignment with European standards on deterrence
One of the most striking aspects of the Draft Bill is its penalty regime. Spain adopts the ranges set out in the AI Act, including penalties of up to 35 million euros or 7 per cent of global turnover.
Beyond the amount, the design of the penalty system incorporates several key elements:
- A detailed classification of minor, serious and very serious infringements is established, with further differentiation according to the type of operator (provider, deployer, etc.).
- It introduces the possibility of complaints, including anonymous ones, which strengthens detection mechanisms.
- Provisional measures are authorised whilst cases are being processed to prevent immediate harm.
- It provides for an obligation to restore the situation to its previous state and to compensate for damages, thereby introducing a restorative dimension alongside the punitive one.
From a compliance perspective, this framework reinforces the need for robust internal control systems, particularly in organisations operating with high-risk systems.
Public administration: towards algorithmic transparency
The Draft Bill devotes a specific chapter to the use of artificial intelligence within the General State Administration. Here, there is a clear focus on transparency and accountability. Public authorities must report on the use of AI systems, contributing to the creation of an interoperable inventory. Furthermore, the role of the ‘artificial intelligence officer’ is introduced, responsible for internal policies and regulatory compliance. This element is particularly significant, as it foreshadows a possible shift towards organisational structures similar to those already in place for data protection (DPOs), adapted to the field of AI. The CES, however, highlights the need to ensure greater consistency with other regulations, such as legislation on equality and non-discrimination, particularly with regard to the assessment of algorithmic bias.
Shortcomings identified: social dialogue and participation
Beyond the technical content, the opinion raises a significant criticism of the regulatory drafting process: the absence of structured social dialogue. Given that artificial intelligence has a direct impact on employment, labour relations and the organisation of work, the ESC considers that greater involvement of trade unions and employers’ organisations in the early stages would have been desirable. This is no minor point. The governance of AI is not merely a technical or legal issue, but also a socio-economic one. The legitimacy and effectiveness of the legislation will depend to a large extent on its acceptance by the stakeholders concerned.
A ‘second-tier’ law with strategic implications
The Draft Bill on the proper use and governance of artificial intelligence is not a disruptive piece of legislation in substantive terms. It does not redefine the principles or requirements of the European system. However, its importance is structural.
It defines who exercises control, how control is exercised, and what happens in the event of non-compliance. In other words, it transforms a European framework into a national operational system. The real challenge will lie not so much in the adoption of the legislation as in its effective implementation: coordination between authorities, technical capacity for supervision, adaptation by organisations and, above all, the integration of a culture of compliance within a field – that of artificial intelligence – characterised by its rapid evolution and technical complexity.
