On 23 September 2026 the Spanish data protection authority (AEPD) published a warning, AI-00009-2026, addressed to a company that was going to use an AI tool to analyse CVs, score applications and rank them in recruitment and internal mobility. The tool had not been deployed. The company had already told the legal representatives of its staff that it was coming. It received the warning anyway.
That is the news: the AEPD acted before a single applicant had been scored.
What a warning is, and what it is not
Article 58(2) GDPR gives supervisory authorities a set of corrective powers. The first, in point (a), is to issue warnings to a controller or processor that intended processing operations are likely to infringe the Regulation.
The warning explains its own nature. Paraphrasing the AEPD's text, which is in Spanish only: its purpose is not to penalise the controller's conduct but to draw the controller's attention to the fact that certain processing could infringe the Regulation, without any certainty being needed that it does.
It is not a penalty and it does not find an infringement. Nor is it a formality without consequences: it opens a named file, puts in writing what the authority expects, and ends by noting that, if the company does not take the necessary measures, investigation or corrective proceedings could follow.
What the company had done
According to the facts set out in the warning, the company had done three things that sound like due diligence:
- Informed the legal representatives of its workers about the forthcoming deployment, in December 2025, under Article 64.4 of the Estatuto de los Trabajadores (Spain's Workers' Statute).
- Presented the tool as support: the system would give a score of fit for the role, and the final screening or rejection decision would rest with a person.
- Stated that the system excluded sensitive attributes and was subject to periodic audits.
The tool was also being developed centrally by the corporate group, and the Spanish company undertook to analyse the group's results before putting it into operation.
What the AEPD asked for
The warning does not prohibit the tool. It says expressly that the GDPR does not prevent the use of these technologies in recruitment, but requires the processing to carry safeguards appropriate to its risks. And it sets them out:
- Data protection by design. Under Article 25 GDPR, which, in the AEPD's words, requires data protection principles and safeguards to be built in effectively both when the means of processing are determined and while it is carried out.
- Assess the risks and, if the processing is high-risk, carry out the impact assessment beforehand. The warning says that in those cases a data protection impact assessment must be carried out, before the processing, on the terms of Article 35 GDPR.
- Transparency towards applicants and workers, based on Articles 12 to 14 GDPR: clear information on the purposes of the processing and on the role the tool plays in the evaluation.
- Effective human intervention. In the AEPD's words, human intervention must be effective, so that the person can assess critically the information or score provided by the system and take the decision without being determined in practice by the automated result.
The fourth point does not find that there is an automated decision within the meaning of Article 22 GDPR. It says that whether Article 22 actually applies must be assessed according to how the decision-making process really works. What counts is how the process operates, not how it is described.
Informing is not authorising
The company had informed the legal representatives, and that did not prevent the warning. The lesson also applies to the individual information required by Real Decreto 723/2026, in force since 5 October 2026: its Article 3(2)(k) requires each worker to be informed in writing of the existence of algorithmic or automated decision-making systems and, where they decide on working conditions such as career progression, of their guidelines, criteria and operating rules.
That document is necessary, and this guide explains how to write it, with a downloadable template. But informing people that a system exists does not answer the other questions: whether the processing is necessary and proportionate, whether its impact was assessed, whether the person reviewing the score can really depart from it. And the royal decree speaks of workers; for someone who is still an applicant, the transparency the warning cites is that of Articles 12 to 14 GDPR.
In its note of 7 October 2026 on algorithmic systems in the workplace, which reads RD 723/2026 and this warning together, the AEPD adds two points worth keeping in mind when documenting. On audits: mentioning periodic audits is not enough, on its own, to demonstrate compliance; to be verifiable, their purpose, their scope and the version of the system examined must be on record. On sensitive attributes: saying that a tool does not use them is not enough; it must be checked whether other variables allow them to be inferred or act as proxies. That is the supervisory authority's reading, not a legal text.
Which documents evidence the safeguards
The warning asks for measures to ensure, and to be able to demonstrate, that the processing complies with data protection law. What can be demonstrated is what is written down, dated before go-live. From a governance standpoint, five pieces carry it:
- The inventory. The tool, what it is used for (recruitment, internal mobility or both), what data it processes and who is accountable for it. It is what lets you say which processing exists before discussing its safeguards.
- Due diligence on the provider. Which variables the system uses, which audits have been carried out, with what scope and on which version, and what prevents an apparently neutral variable from acting as a stand-in for a sensitive attribute. The AEPD asks for this to be checked, not taken on trust. If the provider is the group itself, the note is clear: developing or evaluating the tool centrally does not relieve the entity that decides to use it in Spain of its responsibility.
- The impact assessment, where the processing is likely to result in a high risk, carried out before the processing and with the minimum content of Article 35(7) GDPR.
- Information to the people affected: that of Articles 12 to 14 GDPR to applicants and workers and, for staff, that of Article 3(2)(k) of RD 723/2026.
- The decision log. Who reviewed each score, what they saw, what they decided and when they departed from the recommendation. The AEPD's note suggests recording discrepancies: a systematic absence of them can indicate practical dependence on the score. It is the only way to show that human intervention is effective and not a signature.
Where the same system is also used to evaluate staff performance, two further laws come in; they are brought together in a single table.
What this piece does not cover
Regulation (EU) 2024/1689. The warning does not cite it: its legal basis is the GDPR and the LOPDGDD (Spain's data protection act). The fact that a recruitment system falls within Annex III, point 4, of the AI Act is another layer, with its own timeline, covered in this article on AI in recruitment.
The date of the decision. The published PDF does not carry one. The dates on record are the receipt of the submission by the AEPD, on 6 January 2026, the decision to open investigations, on 16 January 2026, and publication, on 23 September 2026.
The employment law of recruitment. What may or may not be asked of an applicant, or how deployment is negotiated with the workers' legal representatives, are employment law questions this piece does not address.
Content in accordance with AEPD warning AI-00009-2026, issued under Article 58(2)(a) of Regulation (EU) 2016/679; with Articles 22 and 35 of that Regulation; and with Article 3(2)(k) of Spain's Real Decreto 723/2026. The AEPD's warning and note are a case and guidance, not law.
This article is for information purposes only and does not constitute legal advice.