05/06/2026
Spain takes the plunge: the Draft Bill on the Proper Use and Governance of Artificial Intelligence, and what this really means for businesses
by Iván Martínez, CEO of Intedya
Whilst much of the public debate on artificial intelligence continues to swing between technological enthusiasm and dystopian alarmism, the Member States of the European Union are doing what is required of them: transposing Regulation (EU) 2024/1689, the so-called AI Act, into their national legal systems. Spain has made its move with the Draft Bill on the Proper Use and Governance of Artificial Intelligence, spearheaded by the Ministry for Digital Transformation and the Civil Service through the State Secretariat for Digitalisation and Artificial Intelligence. This is no minor piece of legislation. Nor is it, despite what some headlines suggest, a ‘Spanish AI law’ that duplicates the European Regulation. It is somewhat more nuanced and, for those of us working in compliance, more relevant: it is the piece of legislation that makes the AI Act applicable, enforceable and subject to penalties within Spanish territory. And it introduces new provisions of its own that are worth reading through carefully. I’m going to try to outline what really matters, particularly for businesses, compliance officers, DPOs and senior management who are (or should be) preparing for the transition.
What the Act is (and what it isn’t)
The AI Act is a European regulation that is directly applicable. This means it is already binding, without the need for transposition. However, the Regulation itself delegates three critical matters to Member States: the system of penalties, the designation of competent national authorities and the administrative procedure for enforcing it. The Spanish draft bill does exactly that, whilst adding some elements of its own.
In practical terms, the text:
-
Defines the penalty regime applicable in Spain for breaches of the European Regulation.
-
Regulates the legal framework for authorising the use of ‘real-time’ remote biometric identification systems in public spaces for policing purposes, one of the most sensitive aspects of the AI Act.
-
Designates sector-specific market surveillance authorities, using a decentralised model that already foreshadows coordination issues.
-
Establishes the national governance framework, with the Spanish Agency for the Supervision of Artificial Intelligence (AESIA) at its heart.
-
It introduces a new digital right: the right to disconnect from or withdraw from AI systems that have caused serious incidents.
It is not, therefore, a law about what can or cannot be done with AI (that is set out in the Regulation). It is the law that specifies who oversees compliance, how penalties are imposed, and the cost of non-compliance in Spain.
The penalty regime: figures that cannot be ignored
Here, in no uncertain terms, is the part that will capture the attention of boards of directors. The draft bill classifies infringements into three levels (very serious, serious and minor) and sets out penalty ranges that align with the upper limits set by the European Regulation:
-
Very serious infringements involving prohibited AI practices: from €7,500,001 to €35,000,000 or, if the offender is a company or group, from 2 per cent to 7 per cent of the global turnover for the previous financial year, whichever is higher.
-
Very serious infringements involving high-risk systems: from €7,500,001 to €15,000,000, or from 2 per cent to 3 per cent of global turnover.
-
Serious infringements: from €500,001 to €7,500,000, or from 1% to 2% of global turnover.
-
Minor infringements: from €6,000 to €500,000, or from 0.5% to 1% of global turnover.
A critical point for corporate groups: the Draft Bill expressly states that, where the offending company belongs to a group, the upper limit is calculated on the basis of the group’s turnover, not that of the subsidiary. In other words, the traditional approach of channelling high-risk activities through small subsidiaries to limit exposure to penalties is effectively ruled out from the outset. The message to the market is clear: no corporate structuring will be effective. For SMEs and start-ups, however, the principle of the European Regulation applies: the fine will be the lower of the percentage or the fixed amount. Furthermore, a warning mechanism without a fine is introduced for minor infringements by SMEs, subject to the adoption of corrective measures and, where applicable, compensation for the damage caused. This is a soft-landing approach designed not to stifle innovation, and one that should be put to good use.
Decentralised governance: opportunity and risk
Perhaps the most structural aspect of the Draft Bill, and the one that will have the greatest impact on the day-to-day running of businesses, is the model for market surveillance authorities. Spain has opted for a decentralised, sector-based scheme:
-
AESIA as the single point of contact and the authority for prohibited AI systems (in general), high-risk systems listed in Annex III relating to biometrics, critical infrastructure, education and training, employment, essential services, and systems not classified as prohibited or high-risk when they fail to comply with transparency obligations.
-
The Spanish Data Protection Agency (AEPD) and regional data protection authorities for AI systems in biometrics used for policing or border management, ensuring compliance with the law, and migration/asylum.
-
The Bank of Spain and the National Securities Market Commission (CNMV) for solvency assessment and credit rating systems.
-
Directorate-General for Insurance and Pension Funds for systems used in risk assessment and pricing in life and health insurance.
-
The General Council of the Judiciary (CGPJ) and the Central Electoral Board for the areas of the administration of justice and democratic processes.
-
The sectoral authorities designated under the harmonised acts listed in Annex I of the Regulation, for AI systems integrated into regulated products.
This has both a positive aspect (each sector is supervised by those with knowledge of the business) and a cause for concern: the risk of inconsistent application. Similar conduct could receive different responses from different authorities. The Draft Bill recognises the problem and establishes the Joint Commission for the Coordination of Market Surveillance Authorities, chaired and managed by AESIA, which will be tasked with preventing divergent interpretations. It remains to be seen how this will work in practice. For companies with cross-functional AI systems (think of a financial institution using AI for credit scoring, HR management and customer service simultaneously), the landscape of supervisory bodies can become complex. Identifying which authority is competent for each system will no longer be a theoretical matter.
The new right to disconnect: Spain innovates once again in digital rights
One of the most original elements of the Draft Bill is the introduction of a new digital right to disconnect or withdraw from the market AI systems that have caused serious incidents. The explanatory memorandum compares this, not without a certain historical perspective, to the CJEU’s recognition in its day of the ‘right to be forgotten’. What does this entail? That any natural or legal person may report, even anonymously via an external mailbox managed by AESIA, facts that may constitute an infringement, and that the competent authority may adopt precautionary measures, including the withdrawal of the product or the disconnection or prohibition of the AI system within its territorial jurisdiction where the continued operation of the system poses an unacceptable risk.
Taken at face value, this is an extremely powerful tool. For citizens, it provides a genuine mechanism for responding to serious harm caused by AI (the text specifically cites the case of the death of a family member). For businesses, it represents a paradigm shift: mere formal compliance is no longer sufficient; a serious incident may result in the precautionary disconnection of the system even before the disciplinary proceedings have been concluded.
The obligation to report serious incidents – classified as a very serious offence if omitted – is no longer merely a bureaucratic formality. It becomes the linchpin around which the response of the penalty system revolves.
‘Real-time’ remote biometric identification: striking a delicate balance
Chapter III of the Draft Bill regulates one of the most hotly debated aspects of the AI Act: the exceptional use of remote real-time biometric identification in public spaces for policing purposes. Spain has opted for a model of prior judicial authorisation granted by the Administrative Courts, with a reasoned decision issued within 48 hours and tacit rejection in the absence of a response. The circumstances in which authorisation may be granted, set out in Annex I, are exhaustive: searching for victims of kidnapping or human trafficking, preventing terrorist threats or threats to life, and locating suspects of serious crimes (terrorism, human trafficking, sexual exploitation of minors, arms trafficking, murder, rape, sabotage, amongst others).
The framework, with its requirements for proportionality, justification and temporal, geographical and personal limitations, is well designed. The key issue will be its effective implementation and, above all, the courts’ capacity to respond with the speed and technical rigour required for genuine judicial oversight of the police’s use of biometric AI.
Timetable: what is currently in force and what is to come
The Draft Bill provides for a phased entry into force in line with the AI Act’s timetable:
-
2 August 2025: penalty regime applicable to prohibited AI systems and notified bodies.
-
2 August 2026: penalty regime applicable to operators of high-risk AI systems listed in Annex III.
-
2 August 2027: penalty regime applicable to operators of high-risk AI systems listed in Annex I (those integrated into products subject to harmonised Union legislation).
This phased approach is important. Anyone currently operating a system classifiable as high-risk under Annex III has precisely the period from now until August 2026 to bring their operations into line. That is not much time when you consider what is involved: quality management, technical documentation, a risk management system, post-market surveillance, conformity assessment, CE marking, registration in the European database and, for public sector bodies and public service providers, an assessment of the impact on fundamental rights.
What companies should already be doing
From a compliance officer’s perspective, there are a handful of actions that cannot be put off any longer.
-
Firstly, an inventory. You cannot manage what you do not know. Identify all AI systems in use (in-house, embedded in third-party products, in pilot trials) and classify them according to the categories set out in the Regulation (prohibited, high risk, limited risk, minimal risk).
-
Secondly, a role analysis. Determine for each system whether the organisation acts as a provider, deployer, importer, distributor or authorised representative. The obligations – and therefore the infringements – vary substantially depending on the role.
-
Third, gap analysis. Compare the current status against the requirements of the Regulation and the Draft Bill. Pay particular attention to technical documentation, the quality management system, post-market surveillance, transparency towards individuals, and specific obligations in workplace settings (informing workers’ representatives).
-
Fourth, transition plan. A roadmap up to August 2026 setting out those responsible, deadlines, budgets and metrics. If GDPR compliance has taught us anything, it is that regulatory transitions cannot be improvised in the six months leading up to the implementation date.
-
Fifth, coordination with existing frameworks. The AI Act does not operate in a vacuum: it overlaps with the GDPR, with product safety regulations, and with sector-specific regulations. Organisations that already have mature management systems (ISO 27001, ISO 37301, ISO 42001) have a head start and should capitalise on it.
Spain is not breaking new ground with this draft bill: it is fulfilling an obligation imposed by EU law. However, it does so with elements of its own (the right to disconnect, the specific regime for SMEs, the anonymous reporting channel, and the granting of powers to administrative courts to authorise the police’s use of biometrics) which deserve recognition. The fundamental question, however, is another: are Spanish and Latin American companies operating in the European market truly prepared? My impression, after months of working with organisations on both sides of the Atlantic on regulatory compliance transitions, is that the gap is significant. Too many organisations still view AI as a technical matter that can be delegated to the IT department, when in fact – and from 2026 onwards this will be beyond dispute – it is a matter of corporate governance and regulatory compliance, with penalties comparable to, or even greater than, those under the GDPR.
