The Government publishes an ethical guide on AI in the Administration that exempts AAPP from economic fines

The new Government guide sets out the ethical use of AI in the Administration, exempts AAPP from fines, and strengthens training and supervision.

3 minutes

fotonoticia 20261119115746 1920
Add DEMÓCRATA to Google

Published

3 minutes

Spanish Public Administrations will be exempt from financial penalties if they violate the European Artificial Intelligence Regulation (AI Act). However, supervisory authorities may issue warnings and order corrective measures. In return, the Public Administrations must train their staff in this technology, carry out impact assessments on fundamental rights, and inform citizens when it is associated with automated systems or with synthetic content.

These obligations are set out in the "Guide for the Ethical Adoption of Artificial Intelligence in Public Administrations", prepared by the Ministry for Digital Transformation and Public Function through the State Secretariat for Digitalization and Artificial Intelligence. The document, funded with European Union (EU) funds from "NextGenerationEU", is intended as a practical instrument for a "homogeneous, safe and reliable" deployment of AI in the public sector, facilitating compliance with the European framework "without adding new administrative burdens".

The guide adjusts its criteria to the European AI Regulation, whose deadlines were modified on 8 July 2026 by EU Regulation 2026/1744, known as the "AI Omnibus". According to this timetable, the adaptation of the "existing high-risk AI systems intended to be used by public authorities" will be mandatory from 2 August 2030.

Basic principles and uses excluded from the Regulation

The text states that any deployment regulated in the guide is based on four "non-waivable" principles: artificial intelligence does not replace human decision-making — people retain responsibility, professional judgment, and the final administrative ruling —; it may not entail a reduction in rights or guarantees; it must serve to improve public services and not be limited to automating them; and citizens will retain channels for complaint and oversight.

It also specifies which systems are outside the scope of the Regulation: those "used exclusively for military purposes, for defense or for national security", those "whose sole purpose is R&D", those "distributed with free licenses or open source" (to which a "reduced application" applies) and traditional software based on "fixed rules programmed by the human being" or on "simple prediction systems".

On the contrary, the prohibition of uses that imply an “unacceptable risk” is recalled, for conflicting with the values of the European Union. These include “cognitive-behavioral manipulation”, “social scoring and/or social classification”, “emotion recognition in the workplace and in educational settings, except for medical or safety reasons”, the “non-selective extraction of facial images” to create databases, or “predictive police surveillance” based solely on profiling.

Allocation of oversight and effects on public procurement

With regard to national governance, the guide refers to the proposal for the Organic Law Project for the proper use and Governance of Artificial Intelligence, which assigns to the Agencia Española de Supervisión de Inteligencia Artificial (AESIA) the role of “Single Point of Contact” and to the Dirección General de Inteligencia Artificial the status of “notifying authority”.

Market monitoring will be distributed among several bodies: the Agencia Española de Protección de Datos will handle biometrics and border control; the Consejo General del Poder Judicial, the systems used in the Administration of justice; the Banco de España and the CNMV, the assessment of solvency; and the Dirección General de Seguros y Fondos de Pensiones, life and health insurance.

In relation to the exemption of fines for public bodies, the document clarifies that this provision appears “in the proposal for the Spanish Organic Law Project”. In the event of non-compliance, the competent authority will issue a resolution with “a warning” and “the necessary corrective measures to stop the unlawful conduct or mitigate its effects”.

In the procurement field, the Ministry suggests first considering whether the problem “can be solved in a simpler and safer way using other technologies, such as automation”, and highlights current limitations of AI, including the “biases” derived from historical data, its “limited accuracy”, and the “hallucinations” of generative models.

The text also warns that, if an Administration purchases a low-risk commercial solution and carries out “a substantial modification” or changes “its intended purpose” to apply it to essential services —such as the “allocation of social assistance resources”—, it will become legally considered a “provider” of a high-risk system, assuming all technical obligations, CE marking, and conformity assessment requirements.

To improve bidding results, the guide proposes creating “multidisciplinary teams,” checking the availability, quality, and privacy of the data, considering both the “upfront costs” and the “ongoing maintenance costs,” and introducing model clauses that ensure interoperability and energy sustainability.

The document concludes with examples of responsible deployment —such as assistants for processing benefits, emergency management tools, or “chatbots” for citizen service— and a specific annex devoted to models of “general-purpose artificial intelligence”.