Skip to content
Guide

AI and the GDPR, ENS and EU AI Act: what a company using AI must do

Updated
By
The MagnolIA team
On this page
17 min read

Short answer

A company that uses AI on personal data answers for it under the GDPR. The EU AI Act adds obligations graded by risk, and Spain's ENS binds the Spanish public sector and its suppliers. Running AI in your own network does not exempt you from any of them, but it does avoid sending documents to an outside provider. MagnolIA is designed to fit all three.

What does the GDPR require when AI is used on personal data?

The General Data Protection Regulation does not talk about AI: it talks about processing personal data, and pasting a document with customer data into an AI is processing it. These are the articles that weigh most when choosing how to use it.

GDPR articles that affect the use of AI
ArticleWhat it saysWhat it means when using AI
Art. 5(1)(c) · Data minimisationData must be adequate, relevant and limited to what is necessary.The AI should not receive more personal data than the task needs.
Art. 6 · Lawful basisProcessing is lawful only if it rests on one of six grounds: consent, contract, legal obligation, vital interests, public interest or legitimate interests.Using AI does not create a new basis: it is worth checking that the one already covering the processing (the contract with the client, for instance) covers this use too.
Art. 9 · Special categoriesProhibits, with exceptions, processing data such as health data, biometric data or data revealing political opinions or trade union membership.Medical records, medical reports or expert assessments call for far more caution.
Arts. 13 and 14 · InformationPeople must be told the purposes of the processing, its legal basis and who receives their data.If an AI provider receives the data, it is one more recipient your privacy notice should reflect.
Art. 22 · Automated decisionsGives people the right not to be subject to a decision based solely on automated processing that produces legal effects or similarly significantly affects them.AI can prepare the decision; letting it decide alone about a person comes with conditions of its own.
Art. 25 · By design and by defaultAsks for measures such as pseudonymisation to be applied both when the means of processing are chosen and during the processing itself.Where the model runs and what data it receives is decided before anyone starts using it.
Art. 28 · ProcessorIf a third party processes data on your behalf, it must offer sufficient guarantees and be governed by a contract or other legal act.A cloud AI provider that processes your documents is a processor: that contract is needed.
Art. 30 · Records of processingEach controller keeps a record of its processing activities, with purposes, categories of data and recipients. Paragraph 5 exempts organisations with fewer than 250 people, subject to conditions.A settled use of AI with personal data should be findable in that record.
Art. 32 · SecurityRequires technical and organisational measures appropriate to the risk; it cites pseudonymisation and encryption.Access control, encryption and, where appropriate, anonymising documents before processing them.
Art. 35 · Impact assessmentRequires an impact assessment before processing that is likely to result in a high risk, in particular when using new technologies.Depending on the use, an impact assessment may be needed before deploying AI.
Art. 44 · TransfersAllows data to be transferred to a third country only if the conditions of Chapter V are met.You need to know in which country the provider processes the data.
A summary of the articles cited; the full text is on EUR-Lex, linked in the sources.
Back to contents

Who answers for the data: the company using the AI or the provider?

The GDPR calls whoever decides the purposes and means of processing the controller, and whoever processes data on the controller's behalf the processor (Article 4(7) and 4(8)). A company that decides to use AI on its clients' or employees' data is the controller, wherever the model runs. What changes with each way of using AI is who else is involved.

Who is involved in the processing, by way of using AI
Way of using AIWho else processes the dataWhat to have settled
An employee's personal or free account on a public AI toolThe tool's provider, on terms the employee accepted, not the company.This is the uncontrolled case: the company answers for data that left without any contract of its own. House rules and an approved alternative are how it is avoided.
Cloud AI contracted by the companyThe provider, normally as a processor, and the companies it relies on.The Article 28 contract, which country the data is processed in, how long it is kept and whether it is used to train models.
On-premise AI built and run by the company itselfIn principle, nobody else: the document is processed on the company's own machine.Security measures, backups and access are in-house work.
On-premise AI managed by a providerThe document is processed in your network; the provider maintains the machine.What the provider sees while maintaining it, and under which contract. Its role under data protection law is for your data protection officer to assess.
General orientation, not a legal classification: each party's role depends on the contract and on what it actually does with the data. The questions to put to any provider are in what to ask an AI provider.
Back to contents

Can an AI make decisions about people?

Article 22 of the GDPR gives every person the right not to be subject to a decision “based solely on automated processing” which produces legal effects concerning them or similarly significantly affects them. It allows three exceptions: the decision is necessary for a contract, it is authorised by law, or there is explicit consent. In the first and third, the person keeps at least the right to obtain human intervention, to express their point of view and to contest the decision.

Summarising a file, drafting a letter or finding a clause is not deciding about anyone. Rejecting a CV, refusing an application or setting a premium with no person reviewing it may well be. The prudent way to work is for the AI to prepare and for a person with judgement and authority to decide, with a record of who decided.

Back to contents

When is a data protection impact assessment needed before using AI?

Not always. Article 35 of the GDPR requires one where a type of processing, “in particular using new technologies”, is likely to result in a high risk to people's rights and freedoms, and it has to be done before the processing starts. The article itself names three cases in which it is required:

  • Systematic and extensive evaluation of people based on automated processing, such as profiling, where decisions with legal or similar effects are based on it.
  • Large-scale processing of special categories of data, such as health data, or of data on criminal convictions and offences.
  • Systematic monitoring of a publicly accessible area on a large scale.

Outside those cases, the risk of each use decides, and paragraph 4 has each supervisory authority publish a list of the processing that requires one: in Spain, it is worth checking the AEPD's. An office that summarises contracts with AI inside its own network is not in the same position as a clinic processing patient records every day, or a company scoring job applicants. Where there is a data protection officer, the same article provides for the controller to seek their advice when carrying out the assessment (paragraph 2). Where the model runs does not settle on its own whether one is needed, but it does change what the assessment has to look at: with the AI in your network, the document is not sent to an outside provider's servers and does not leave the EU.

Back to contents

What has Spain's data protection authority published on generative AI?

The Agencia Española de Protección de Datos (AEPD) is Spain's national data protection supervisory authority. It has no single “generative AI for businesses” guide, but three recent documents serve as a reference, each written for a different reader. All three are in Spanish.

The rules the AEPD applies to itself (January 2026)
The summary of obligations and recommendations for managing generative AI at the AEPD is the authority's internal rule, not an instruction to businesses, which is exactly why it is useful: it shows how the regulator organises itself. It asks for use cases involving personal, sensitive or confidential information to be deployed preferably on internal systems, or on ones that keep the organisation in control; for an inventory of the systems in use; for an assessment, before contracting, of what the provider does with metadata, logs and telemetry, whether it uses the data for training, where it stores it and how long it keeps it; and for no AI-generated content to go into documents with legal effects without human review and validation.
The ten tips for the public (27 January 2026)
“Cuidado con lo que le confIAs” (roughly, “mind what you confide in it”) advises against sharing personal data or sensitive information with AI and, at work, tells people to follow their organisation's policies and keep confidential data about the organisation, its staff or its clients out of it.
The guidance on agentic AI (18 February 2026)
Written for controllers and processors who want to use agents in processing personal data: it describes the technology, its vulnerabilities and the measures that can be taken.

The AEPD's February 2020 guide on bringing processing that embeds AI into line with the GDPR is still published; the PDF now carries an “under review” mark. It predates generative AI and is aimed at organisations that build AI components into their processing.

What does the EU AI Act require of those who use AI?

Regulation (EU) 2024/1689, known as the AI Act, uses the term "deployer" for whoever uses an AI system under its own authority. A company that uses AI in its work is one. Its obligations depend on the risk of the use, not on the tool.

  • AI literacy (Article 4). Take measures to support the AI literacy of staff who use AI systems. Since the Omnibus, the rule makes clear that it does not require guaranteeing a specific level for any individual.
  • Transparency (Article 50). The provider must design the system so that people know they are interacting with an AI. The deployer must give notice when, among other cases, it publishes deepfakes or uses emotion recognition or biometric categorisation.
  • High-risk uses (Article 26). Whoever deploys a high-risk system must use it according to its instructions, assign human oversight to people with the competence, training and authority to exercise it, and keep the logs the system generates for at least six months.

Application is staggered, and the simplification package known as the AI Omnibus (Regulation (EU) 2026/1744), in force since 27 July 2026, postponed the high-risk rules and rewrote Article 4. This is the timeline published by the European Commission.

Application timeline of the EU AI Act
DateWhat starts to apply
1 August 2024The Regulation enters into force.
2 February 2025Prohibited AI practices and AI literacy obligations.
2 August 2025Governance rules and obligations for general-purpose AI models.
2 August 2026General application, including the transparency rules.
2 December 2027High-risk rules for uses in areas such as biometrics, critical infrastructure or education (deadline extended by the Omnibus).
2 August 2028High-risk rules for systems embedded in regulated products (deadline extended by the Omnibus).
Source: European Commission, regulatory framework for AI, accessed November 2026. Articles are cited from the consolidated text of 27 July 2026, which incorporates the Omnibus. The detail, date by date and with what the Omnibus changed, is in the EU AI Act timeline.

Who is bound by Spain's National Security Framework (ENS)?

The ENS (Esquema Nacional de Seguridad) is Spain's mandatory security framework for public-sector information systems, set out in Royal Decree 311/2022. It applies to the whole Spanish public sector, and also to private entities that provide services or solutions to public bodies under contract (Article 2).

Each system is assigned a category (basic, medium or high) according to the impact an incident would have. Medium and high-category systems need an audit to certify conformity; basic-category systems need a self-assessment to declare it (Article 38). Which Annex II measures affect an AI system, and what to ask a supplier, is set out in AI and Spain's ENS.

Back to contents

What does on-premise AI solve, and what does it not?

On-premise AI changes the route the data takes. The document is processed on a machine in your network: it is not sent to an outside provider's servers and does not leave the EU. If the service is managed, whoever maintains the machine is still a supplier you sign a contract with; what role that supplier has under data protection law is for your data protection officer to assess.

It does not solve what depends on you. You still need a legal basis for processing the data, and you still have to decide who can access what, train your people, judge whether a use is high-risk and carry out the impact assessment when one is due. No technology complies with a regulation on your behalf.

What MagnolIA contributes and what remains your organisation's job
ObligationHow MagnolIA helpsWhat remains your job
Not handing data to third parties without guaranteesPrompts, documents and answers are processed in your network. MagnolIA only sees the machine's monitoring.Deciding which data is processed with AI, and on what legal basis.
Security of processingTLS-encrypted entry point, role-based access, two-factor authentication for administration and AES-256 encrypted backups.Managing user onboarding and offboarding, and protecting workstations.
Data minimisationThe anonymiser replaces personal data before the document is processed and restores it afterwards.Deciding when to use it, and reviewing what it flags.
Logging and oversightPer-user usage records and a chained, verifiable audit log, kept for two years in your installation.Reviewing those records and assigning human oversight to named people.
Staff trainingSupport.Training staff and setting internal rules of use.

Each measure is described in on-premise AI security. The legal terms are defined in the enterprise AI glossary.

Back to contents

Which mistakes are worth avoiding when using AI with personal data?

  • Assuming nobody uses it. Where there is no approved tool, some staff may be using a personal account. That use goes through no contract of the company's.
  • Reading “business plan” as “problem solved”. The contract governs the processing; it does not change the fact that the document leaves the network, nor does it spare you checking country, retention and subcontractors.
  • Treating pseudonymisation as anonymisation. A document with labels instead of names is still personal data for whoever holds the mapping table. The detail is in anonymising documents with AI.
  • Letting the output be used unchecked. A model can be wrong with complete confidence. If the text goes to a client, into a file or into a decision, a person reviews it.
  • Forgetting people's rights. Access, rectification and erasure still apply to whatever is stored: conversation histories, uploaded documents and logs.
  • Buying the tool before asking the data protection officer. Changing a plan costs less than changing a deployment.

The GDPR provides for fines of up to €20 million or 4% of total worldwide annual turnover, whichever is higher, for the most serious infringements (Article 83(5)). The amount in any given case depends on its circumstances; the figure gives the order of magnitude, nothing more.

Back to contents

Where does a company start if it wants to use AI in line with the GDPR?

  1. Take stock of real use

    Which AI tools are already in use, by whom, for what and with which accounts. Include what nobody approved.

  2. Classify the information

    Separate what holds no personal data, what does, and what is special-category data or under professional secrecy. Each group allows different tools.

  3. Decide where each group is processed

    Cloud under contract, cloud with the document anonymised first, or a machine in your network. The comparison is in on-premise AI vs cloud AI.

  4. Get the paperwork in order

    A processor contract with each provider that handles data, an up-to-date record of processing and a privacy notice that reflects the use.

  5. Assess the risk

    With your data protection officer: whether an impact assessment is needed and whether any use falls into the AI Act's high-risk categories.

  6. Write the rules and train your people

    What may be given to the AI, what may not, and who reviews the output. There is a starting point in the house rules for AI in the office.

  7. Review from time to time

    Uses change faster than documents do. Repeat the stocktake and look at the usage logs.

What should you check before approving a specific use of AI?

Eight questions for the person who signs it off. If one has no answer, the use is not ready yet.

  • What personal data does the AI receive in this use, and is it the minimum the task needs?
  • Which lawful basis covers the processing, and does it cover this use too?
  • On which machine and in which country is the document processed, and who can see it?
  • If a provider is involved, is there a contract, and does it say what happens to the data at the end?
  • Are conversations and documents stored? For how long, and who has access?
  • Who reviews the output before it goes out or before someone decides on it?
  • Does the output affect a specific person: a job, credit, a benefit, their health?
  • Is the use in the record of processing and in the house rules?
Back to contents

Frequently asked questions about AI and compliance

Does on-premise AI help with GDPR compliance?

On-premise AI helps with the GDPR because the document is processed in your network: it is not sent to an outside provider and does not leave the EU. MagnolIA is designed to fit the GDPR, but it does not replace the assessment of each processing activity, which remains the job of your organisation and its data protection officer.

Can you put personal data into a generative AI tool?

Putting personal data into a generative AI tool is processing under the GDPR: it needs a lawful basis, must be limited to what is necessary and, if a provider is involved, needs a contract. Spain's data protection authority advises people at work to follow their organisation's policies. With on-premise AI such as MagnolIA, the document is processed in your network.

Do you need a data protection impact assessment to use AI?

A data protection impact assessment is not needed for every use of AI. Article 35 of the GDPR requires one where the processing is likely to result in a high risk, for example when processing health data on a large scale or evaluating people by automated means. It is done before starting, with the data protection officer's advice.

What does the EU AI Act require from a company that uses AI?

The EU AI Act sets obligations according to the risk of each use: transparency, human oversight, record-keeping and staff AI literacy, among others. Running AI in your own network does not exempt you from them, but it makes the work easier: every use is logged in your installation, with per-user records and a verifiable audit log.

Can an AI make decisions about people under the GDPR?

An AI should not make decisions about a person on its own: Article 22 of the GDPR gives people the right not to be subject to a decision based solely on automated processing that produces legal or similar effects, subject to exceptions with safeguards. The AI can prepare the decision; it is taken by a person with the judgement and authority to change it.

Who is responsible for the data when a company uses AI?

A company that decides to use AI on personal data is, as a rule, the controller, wherever the model runs. A cloud AI provider normally acts as a processor and needs a contract. With on-premise AI the document is processed in the company's own network; the role of whoever maintains the machine is for the data protection officer to assess.

Is MagnolIA certified under Spain's National Security Framework (ENS)?

MagnolIA does not claim any certification under the ENS, the security framework that Spanish public bodies and their suppliers must follow. What it offers are design measures built with the ENS in mind: role-based access, two-factor authentication for administration, an encrypted entry point, an isolated network, a chained, verifiable audit log kept for two years and AES-256 encrypted backups.

Can AI be used with patient health data?

Health data is a special category under Article 9 of the GDPR, so using AI on it means controlling where it is processed. For a client in the insurance sector that works for healthcare professionals, MagnolIA installed a machine in its office: documents containing patient data are read there and the answers are produced there, with no reliance on any external service.

Is anonymising documents enough to comply with the GDPR?

Anonymising is not enough on its own. What MagnolIA Nexus does is pseudonymise (Article 4(5) GDPR): it replaces the data and keeps the information needed to restore it separately, on your machine. It is a technical measure cited in Article 32 and applies data minimisation, but it does not replace assessing each processing activity.

Sources

External figures on this page link to their original source.

  1. EUR-Lex. Regulation (EU) 2016/679, General Data Protection Regulation (GDPR). eur-lex.europa.eu/eli/reg/2016/679/oj

  2. EUR-Lex. Regulation (EU) 2024/1689, Artificial Intelligence Act (text published on 12 July 2024). eur-lex.europa.eu/eli/reg/2024/1689/oj

  3. EUR-Lex. Regulation (EU) 2026/1744 amending the AI Act (Digital Omnibus on AI). Official Journal of 24 July 2026. eur-lex.europa.eu/eli/reg/2026/1744/oj

  4. European Commission. Regulatory framework for AI: application timeline of the AI Act. Accessed November 2026. digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai

  5. BOE (Spain's Official State Gazette). Royal Decree 311/2022 of 3 May regulating the National Security Framework, ENS (consolidated text, in Spanish). www.boe.es/buscar/act.php?id=BOE-A-2022-7191

  6. AEPD (Spain's data protection authority). Summary of obligations and recommendations for managing generative AI at the AEPD (PDF, in Spanish). January 2026. www.aepd.es/guias/sumario-recomendaciones-iag-aepd.pdf

  7. AEPD (Spain's data protection authority). The AEPD publishes ten recommendations for protecting privacy when using AI tools (press release, in Spanish). 27 January 2026. www.aepd.es/prensa-y-comunicacion/notas-de-prensa/aepd-publica-decalogo-recomendaciones-proteger-privacidad-al-usar-ia

  8. AEPD (Spain's data protection authority). Agentic artificial intelligence from a data protection perspective (PDF, version 1.2, in Spanish). February 2026. www.aepd.es/guias/orientaciones-ia-agentica.pdf

  9. AEPD (Spain's data protection authority). Bringing processing that embeds artificial intelligence into line with the GDPR: an introduction (PDF, in Spanish). February 2020. www.aepd.es/guias/adecuacion-rgpd-ia.pdf

Let's look at your case with the rules in hand

Tell us what data you work with and what worries you. We will show you what stays inside your network, what gets logged and what would remain your organisation's job.

Request a pilot