Transparency in high-risk AI: what Guide 8 of the Spanish sandbox recommends
In July 2026 the Ministry for Digital Transformation published version 2.0 of its guide on transparency of high-risk AI systems, adapted to the Digital Omnibus. It is not binding, but it turns Article 13 of the AI Regulation into concrete measures of interest to those who advise on or procure these systems.
The Spanish pilot artificial intelligence regulatory sandbox, which we analysed in the article on the AI sandbox under RD 817/2023, has produced a series of technical guides. One of them, Guide 8, "Transparency and provision of information to users", was published in its version 2.0 on 17 July 2026, adapted to the changes of the Digital Omnibus. Version 1.0 dates from 10 December 2025.
The guide is credited to the Ministry for Digital Transformation and the Civil Service, under a Creative Commons licence, and is available on the website of the Spanish Agency for the Supervision of Artificial Intelligence.
What it is and what it is not
The guide itself clarifies in its preamble: it is not binding and neither replaces nor develops the applicable rules. It offers practical recommendations while harmonised standards are approved. For a firm, that means it is not a source of obligations, but it is a useful reference for interpreting how compliance is expected to look.
Article 13 of the AI Regulation
The guide focuses on Article 13 of Regulation (EU) 2024/1689, on transparency and the provision of information to deployers of high-risk systems:
- Paragraph 1 requires that these systems be designed so that their outputs can be interpreted and used correctly.
- Paragraph 2 requires instructions for use.
- Paragraph 3 lists the minimum information those instructions must contain.
The guide links this article with others in the Regulation, such as Articles 9, 11, 12, 14 and 15. The timetable for applying the high-risk obligations, which the Digital Omnibus amended, is explained in our analysis of the new AI Regulation timetable.
Who must do what
| Party | What falls to them, according to the guide |
|---|---|
| Provider | Design the system with sufficient transparency and draw up instructions for use with their contact details, characteristics, capabilities and limitations, changes, human oversight, resources and maintenance, and logging mechanisms |
| Deployer | Interpret and use the outputs correctly, apply the instructions for use, identify their own transparency requirements according to the use case and define their needs |
The second row is of particular interest to clients who use high-risk systems developed by others.
Some of the measures it proposes
The guide proposes 32 non-binding measures. Among them:
- A point of contact with the provider, with an online channel for requests and incidents.
- Define the functional domain and the purpose of the system, and anticipate reasonably foreseeable misuse.
- Document the data sources.
- Explain the results from the general to the particular, with counterfactual examples: what would have changed the result.
- Adapt the language to each audience: managers, technicians and affected persons.
- Prefer the minimum complexity that meets the objective.
- Define metrics for accuracy, robustness and cybersecurity, with retraining or stop thresholds.
- Distinguish correlation from causation and avoid revealing sensitive or confidential information.
The guide also refers to a self-assessment questionnaire published in a separate document.
Usefulness for advice
In procurement. When a client acquires a high-risk system, the list of minimum information in Article 13.3 and the guide's measures provide a good index of what to demand from the provider in the contract.
In challenges. Whoever defends a person affected by a decision supported by a high-risk system can use the guide to know what documentation should exist and ask for it.
In-house. Even if a firm does not develop high-risk systems, the guide's ideas, such as adapting explanations to the audience or documenting the data, can be transferred to any use of AI.
More analysis in Legal AI.
Fuentes
- Ministerio para la Transformación Digital y de la Función Pública. Guía 8. Transparencia y provisión de información a los usuarios, versión 2.0, 17 de julio de 2026 (PDF). Accessed 18 October 2026.
The specific data in this article come from the official sources cited.
