Summary

A practical GDPR guide for European businesses deploying an AI agent with personal data. Follow the work from its purpose and legal basis through suppliers, rights, transfers, incidents and changes after launch.

What you'll learn

  • How to connect each agent action to the relevant GDPR duty
  • What to decide about lawful basis, suppliers, notices and transfers
  • How to prepare for rights requests, deletion and incidents
  • What evidence and ownership belong in a launch and change review

A European business running an AI agent with personal data still needs a lawful purpose, a legal basis, appropriate safeguards and a way to honour people’s rights. Map the job from collection through deletion, including the suppliers and retained outputs, before the first live run.[1]

In Eurostat’s 2025 data, among EU enterprises with at least 10 employees that had ever considered AI, 70.31% cite missing expertise, 53.61% unclear legal consequences and 52.72% data-protection or privacy concerns as reasons for non-use. These are AI-technology survey responses, not an agent-compliance rate.[5] The wider picture of AI adoption across European enterprises shows the same size gap behind both numbers.

This guide answers the legal-clarity half for one specific case: a European business running an AI agent that touches customer or client data. It maps what the agent does — read, store, send, transfer, retain — to the GDPR obligations each action engages. For the underlying distinction between a model, an agent, and a workflow, start with what an AI agent is. For the separate EU AI Act questions, the EU AI Act deployer guide is the owner.

This guide is not legal advice, not a compliance assessment, and not certification.

Which GDPR principles apply to an AI agent?

Article 5 sets the processing principles. Purpose limitation, minimisation, storage limitation and accountability give you concrete questions about the agent’s access, actions and retained copies; fairness, transparency, accuracy and security remain relevant too.[1]

Purpose limitation (Article 5(1)(b)) requires data to be "collected for specified, explicit and legitimate purposes and not further processed in a manner that is incompatible with those purposes." An agent that rewrites its own plan mid-run can drift past the purpose it was given.[1]

Data minimisation (Article 5(1)(c)) requires data to be "adequate, relevant and limited to what is necessary." A whole-mailbox permission needs justification against the actual purpose. Prefer a smaller record set where it can do the job; the permission’s breadth alone does not settle the legal assessment.[1]

Storage limitation (Article 5(1)(e)) requires data to be kept "for no longer than is necessary." Agents produce derived artifacts — summaries, labels, embeddings — that outlive the task that created them.[1]

Accountability (Article 5(2)) requires the controller to "be responsible for, and be able to demonstrate compliance with" the rest. Demonstrating compliance means producing evidence of what the agent did.[1]

PrincipleGDPR articleWhat an agent changes
Purpose limitation5(1)(b)The agent can extend its own plan past the stated purpose
Data minimisation5(1)(c)Tools grant access at the account level; the agent works at the record level
Storage limitation5(1)(e)Derived artifacts persist after the source task ends
Accountability5(2)Evidence must cover automated actions no person observed

Article 25(2) turns this into a default setting rather than an aspiration. It requires that "by default, only personal data which are necessary for each specific purpose of the processing are processed," and it names four dimensions: "the amount of personal data collected, the extent of their processing, the period of their storage and their accessibility."[1]

Those four dimensions are the same four levers you set when configuring an agent's permissions.

Who is the controller when an AI agent acts?

A business that decides the purposes and essential means of personal-data processing is its controller. Configuring an agent often expresses those decisions: whose data is involved, what the job is, who receives information and how long it remains. Assess the actual arrangement, including any joint decisions with another organisation.[1][2]

The EDPB’s Guidelines 07/2020 treat controller and processor as functional roles. A contract is relevant, but the factual activities determine the role. Paragraph 40 identifies data categories, duration, recipients and affected people among the essential means reserved to the controller.[2]

Read those questions against the proposed workflow:

  • Which sources and people are in scope?
  • What may the agent do with their information?
  • Which services receive a copy or an output?
  • How long do the source, memory and logs remain?
GDPR still applies when an agent does the work. The controls have to move to runtime.

Spain’s AEPD addresses agents directly in its February 2026 guidance. It treats the agent’s autonomy as a controller design decision and connects minimisation to the system’s configuration. Its guidance helps frame the design review; the legal assessment remains about the actual processing.[3]

Say the business asks a service provider to draft replies from one customer queue. The provider has discretion over implementation, but that does not by itself make it the controller. If it separately uses the messages for its own purpose, assess that distinct processing and role. Article 28(10) addresses a processor determining purposes and means contrary to the Regulation.[1][2]

Navaie’s IAPP opinion proposes mapping roles to actual service use. Use that as an operational reminder to keep the approved arrangement and the observed data flows together, rather than assuming that every change of tool changes the legal role.[4]

The same connected tool shown twice: as processor when acting on instruction, and as an independent controller when it determines its own purpose, with the deploying business positioned as controller across both.
Assess roles for the actual processing. A provider’s distinct own purpose needs a separate role review; tool selection alone does not change a role.

A signed data processing agreement records an intended arrangement. It does not determine what the agent did at 03:00 on a Tuesday.

Why must an agent’s assessed workflow match its actual work?

Traditional compliance describes processing once, in advance, in a document. That model assumes the processing does not change between reviews. An agent breaks that assumption in three specific ways.[3]

The plan changes mid-run. An agent given a goal selects its own steps. Compare the chosen steps with the assessed operation and controls; an unexpected step may require a new review.[3]

Tools get called that no assessment listed. An agent permitted to use a translation or summarisation service may choose it during a task. Each call is potentially a disclosure to a recipient, and possibly a transfer under Chapter 5.[3]

Derived data outlives its source. Summaries, labels and stored search representations can remain after a task closes. The AEPD highlights personal data in agent memory and logs. Include those copies in retention and rights procedures.[3]

Navaie illustrates the issue with a scheduling agent that reads a discharge note, adds a health-related label, uses outside services and retains derived information. It is the author’s hypothetical scenario, not an enforcement case.[4] The practical review needs to cover each of those operations rather than only the calendar entry.

Article 9 sets additional conditions for special-category information, including health data. In this example the discharge note already contains health information, and the additional label creates another record to assess. Ordinary customer contact details do not become a special category merely because an agent reads them.[1]

Before a scheduling workflow sees an attachment, decide whether it needs that attachment at all. If it encounters unexpected health information, provide a route to stop or obtain qualified review rather than letting the calendar task silently expand into classification of a person.

Which GDPR duties follow from each agent action?

Each thing an agent does engages a different set of articles. This is the map, and it is the part worth keeping.[1]

Agent actionPrimary articles engagedControl that answers it
Reads a mailbox, file store, or CRM5(1)(b) purpose, 5(1)(c) minimisation, 25(2) by defaultScope access to a label, folder, or record set — not the account
Stores a summary, label, or embedding5(1)(e) storage limitation, 25(2) storage periodRetention rule per memory tier, with deletion that runs on schedule
Sends an external message5(1)(a) transparency, 22 where the decision is significantHuman approval before external sends
Discloses data to a third-party service30(1) recipient records; 28 if the recipient is a processor; Chapter 5 when the international-transfer criteria are metNamed recipients and roles, applicable transfer arrangements, and a record of each call
Retains logs and traces5(2) accountability, 30(1) records, 32 securityAction log with a defined retention period and restricted access
Infers a new attribute about a person9 where the inference is special-categoryBlock or route inference beyond the stated purpose
Six agent actions on the left — read, store, send, transfer, retain, infer — each connected to the GDPR articles it engages and then to the control that answers it.
Start from what the agent does. The obligations follow the action, not the tool category.

Is Article 22 as broad as it is usually described?

Article 22 is narrower than it is usually described. It gives a data subject "the right not to be subject to a decision based solely on automated processing, including profiling, which produces legal effects concerning him or her or similarly significantly affects him or her." The AEPD distinguishes general automation from decisions within Article 22.[3] Routine drafting or filing will often differ from a decision with significant effects. Assess the actual role of the output, particularly if a person merely confirms an automatically produced rejection.[1][3] Where Article 22 does apply, Article 22(3) requires safeguards including the right to obtain human intervention, to express a point of view, and to contest the decision — a generic approval button does not by itself implement all those rights. Paragraph 3 applies to the contract and explicit-consent cases in Article 22(2)(a) and (c); the legally authorised case requires the safeguards set by the relevant law.[1]

Where the human sits is a design choice, and the AEPD sets out four levels of agent autonomy:[3]

Autonomy levelWho actsPractical effect
The agent proposes, the human operatesHumanThe human executes the proposed action; the agent may still read information
The agent and the human collaborateBothShared execution, shared judgement
The agent operates, the human is consulted or approvedAgent, gatedPrior review; a separate assessment of applicable legal safeguards remains necessary
The agent operates, the human observesAgentOversight is after the fact, not before

Choosing a level is not a technical preference. It decides whether a person can intervene before an action lands.[1]

Does the Article 30(5) exemption cover a small business running an agent?

Article 30(5) rarely rescues a small business running an agent. The exemption applies to organisations with fewer than 250 employees, but it fails if any one of three conditions is met: the processing is likely to result in a risk to rights and freedoms, the processing is not occasional, or it includes Article 9 special-category or Article 10 criminal-offence data. The middle condition is the one that bites. An agent that checks a mailbox every fifteen minutes is difficult to describe as occasional processing.[1]

The agent's goal is a first-class object. When it proposes to expand what it touches — a new tool, a new data category, a new recipient — the platform must surface that change and route it to a human or block it. A static approval given at launch cannot carry an autonomous system through the decisions it makes afterwards.

Where the exemption fails, Article 30(1) applies in full: controller and relevant contact details, purposes, data-subject and data categories, recipient categories and applicable third-country transfers; where possible, erasure time limits and a general security-measures description.[1]

Which controls help put GDPR decisions into practice?

Navaie proposes four build-once controls for agentic systems. They are an expert proposal rather than a legal requirement, and they map cleanly onto the obligations above.[4]

ControlWhat it doesObligation it serves
Purpose locks and goal-change gatesMakes the approved goal visible and blocks or escalates when scope expands5(1)(b), 25(2)
End-to-end execution recordsRecords the plan, actions, data categories, destinations and changes5(2), 30(1)
Tiered memory governanceGives temporary context and retained profiles separate retention rules5(1)(e), 25(2)
Live controller and processor mappingKeeps observed data flows, assessed roles, terms and transfer arrangements together28, Chapter 5, 30(1)
Four labelled control layers sitting between an agent and its connected tools: purpose lock, execution trace, memory tiers, and role registry.
Each control produces the evidence one obligation asks for.

In Yardwork implementations, the equivalent controls are set during scoping rather than added afterwards. The documented practice is scoped tool access per workflow, per-client isolation and permission boundaries, an audit log of actions, approval flows for high-impact actions, and human review before external sends, posts, writes, or commits. Those controls are scoped per engagement. They are a description of implementation practice, not a compliance guarantee, an audit, or certification.

The broader control design — permissions, change control, failure review, scope limits — belongs to the AI agent governance owner. For the mechanics of the approval step itself, what approval workflows do in an AI agent system covers the gate, the timeout, and the fallback.

What should you ask the provider before launch?

The useful provider questions are the ones whose answers become evidence. Vague reassurance is not evidence.[1]

On roles. For each connected tool and model service, are you acting as our processor, and on what instruction? Which calls, if any, do you make for your own purposes? A provider that cannot separate these is describing Article 28(10) risk.[1]

On recipients and transfers. Name every third party that receives personal data during a run, including model, summarisation, translation, and enrichment services. In which country is each processed? What is the transfer basis?[1]

On retention. How long is each memory tier kept — working context, conversation history, embeddings, logs? Does deletion remove derived artifacts, or only the source record?[1]

On evidence. Can you produce, for a specific date, what the agent did, which data categories it touched, which tools it called, and what it sent? Action records can help demonstrate accountability; Article 30’s processing inventory and an individual run log have different jobs.[1]

On revocation. What are the steps to cut an agent's access to one tool, and how long do they take? What happens to data already held?[1]

On security. Article 32(1) requires "appropriate technical and organisational measures to ensure a level of security appropriate to the risk," and names encryption and pseudonymisation, ongoing confidentiality and resilience, restoration after an incident, and "a process for regularly testing, assessing and evaluating the effectiveness" of those measures. Ask which of these apply to the agent's stored credentials and retained context specifically, not to the platform in general.

Which of the tools a team already uses publish their sub-processors and sign a DPA, each with its source and check date: AI sub-processors, plan by plan.

What lawful basis covers the agent’s work?

Identify a lawful basis for each purpose before personal data enters the workflow. Article 6 provides the available bases; permission to install a tool or a supplier’s processing agreement is not one of them. Special-category processing needs an applicable Article 9 condition as well.[1][8]

Write the purpose as work a person can recognise: answering an existing customer’s service question, producing an invoice or screening a job application. “Using AI” describes the means. It does not explain why the information is needed. Keep the purpose narrow enough that a manager can spot when a new request changes it.

The common options need different reasoning:

  • Contract: identify the contract with the individual and why the processing is necessary to perform it or take requested steps before it. A convenient feature is not automatically necessary.
  • Legal obligation: name the applicable EU or national requirement and the information it requires. Do not substitute an internal policy for that law.
  • Legitimate interests: identify the interest, test necessity and balance it against the person’s interests and rights. Record safeguards and reasonable expectations.
  • Consent: check that refusal and withdrawal are real options, with a specific, informed and unambiguous choice. An employee’s agreement needs particular care because of the power imbalance.[8]

Say a customer supplies an address so you can deliver an order. An agent copying that address into the delivery record and an agent selling a profile to another business are different purposes. Review the second proposal independently; a label saying “customer operations” should not let it inherit the first decision.

Record the basis alongside the operation, the responsible person and the reasoning. If a proposed task cannot be justified, restrict the task before connecting the data. Asking a person to approve every output does not supply a missing lawful basis.

What must customers and employees be told?

Give people clear information about your processing, including its purpose, lawful basis, recipients, retention and their rights. Articles 13 and 14 distinguish information collected directly from a person from information obtained elsewhere, with different timing rules and exceptions.[1][9]

The notice should describe what changes for the person. “We use innovative systems” does not tell a customer that messages will be analysed to prepare replies, which service providers receive them or how long the resulting summaries remain. Equally, avoid claiming every message is read by an employee if that is no longer the process.

For each workflow, review:

  • who is responsible and how the person can contact them;
  • which information is used and where indirectly collected information comes from;
  • the purpose and lawful basis, including the interest where applicable;
  • recipient categories, any relevant international transfer and safeguards;
  • retention periods or how they are determined;
  • the applicable rights and how to exercise them;
  • relevant automated-decision information where those provisions apply.[1][9]

Keep this work connected to the actual configuration. Say the initial agent drafts from a customer’s current request. Later it gains access to old correspondence and an external enrichment service. Review the information notice and lawful basis before enabling that expansion. A supplier’s privacy policy describes its own processing; it does not replace your explanation to your customers.

For product-specific questions, the ChatGPT GDPR guide and Microsoft Copilot GDPR guide explain the separate product and account choices. Use this hub to retain the business’s duties across those choices.

What belongs in supplier and subprocessor checks?

Identify each recipient’s actual role and put the required arrangement in place before processing. Article 28 requires sufficient processor guarantees and binding terms covering the instructed work. It also regulates the engagement of further processors; a supplier list is the start of that review.[1][2]

Make the procurement record concrete. Name the contracting entity, the service and account edition, the intended processing, the locations involved and any further processors. Then ask which terms actually cover the proposed use. A consumer account and an organisation’s contracted service should not be treated as interchangeable because the product name matches.

The processor terms should address documented instructions, confidentiality, security, help with rights requests and other relevant duties, deletion or return at the end, and information and audits needed to demonstrate compliance. Record whether further processors need specific or general written authorisation and how intended changes will be notified.[1]

Try a change request before launch. Ask what happens if the supplier replaces the service handling summaries, adds a support location or changes retention. Identify who in your team receives the notice, reviews the consequences and can stop use before a change creates an unresolved issue.

A useful supplier review ends with decisions:

QuestionRecord to keepOperational consequence
Who receives personal data?Named service, entity and roleOnly reviewed destinations are enabled
What instructions apply?Processing terms and agreed scopeExtra purposes need a new review
Who else helps deliver the service?Further processors and change procedureNotices reach a responsible person
How does the service end?Return, deletion and evidence procedureOffboarding includes held copies

Do not infer an independent-controller relationship just because an agent chose a tool during a run. Roles follow the actual purposes and means. Conversely, a contract labelled “processor” should not conceal a supplier’s own distinct use of the information.[2]

Does European hosting settle international transfers?

Map the recipients and access arrangements as well as the storage location. Chapter 5 addresses transfers to third countries or international organisations. The EDPB explains the routes through adequacy decisions, appropriate safeguards and limited derogations; their availability depends on the particular transfer.[1][10]

Begin with the actual service chain: the business system, agent operator, AI service, storage, support and other connected recipients. Record which entity processes what, from where, and under which arrangement. Keep a service’s sales region separate from its processing and support arrangements.

Where an adequacy decision is relevant, check its scope and whether the recipient and processing fit it. Where contractual safeguards are used, identify the applicable arrangement and assess whether it works for that transfer. Do not treat a generic statement that a supplier “uses SCCs” as the completed review.[10]

Say customer messages remain in a European business system, but a connected translation service receives selected passages. The useful question is what happens to those passages, including support access and retained copies. The location of the original mailbox answers only one part of the question.

The data-residency guide covers location and access questions in greater detail. Keep its result in the recipient and transfer record here, and revisit that record when a new service is connected.

How do access, correction and deletion requests reach every copy?

Design a request process that can find personal data across the source system and the agent’s retained outputs. The GDPR rights include access, rectification, erasure, restriction and objection, with conditions and exceptions. The normal response period is one month; a qualifying extension must be communicated within that initial period.[1][9]

Build the search around the person and the processing, rather than a single inbox. Relevant information may sit in customer records, sent messages, saved summaries, search indexes, provider history and operational logs. Identify which systems hold which categories and which team or supplier can act on them.

Use a test person in a controlled exercise. Create a source record, run the workflow, and then follow the record into its outputs. Ask the responsible team to locate it, correct an inaccurate statement and demonstrate the applicable deletion or retention decision. Record failures while the design can still change.

For each request, separate the legal decision from its execution:

  1. Receive the request and confirm identity where needed without collecting unnecessary information.
  2. Determine which right applies and any relevant limits or retention duties.
  3. Locate the information and coordinate with the processors and other relevant recipients.
  4. Carry out the decision in the affected systems and record exceptions or outstanding actions.
  5. Respond clearly, retaining evidence of the decision and the steps taken.

The AI-agent data-erasure guide follows deletion through derived copies and retention arrangements. A button that deletes the original message should not be assumed to remove every summary or copy. Equally, an erasure request does not automatically override every applicable legal retention duty.[1][9]

When do you need a DPIA or a data protection officer?

Assess the processing against Article 35’s high-risk test and applicable authority lists before launch. Separately assess Article 37 and applicable national rules for a data protection officer. These are different requirements; neither follows solely from buying an AI product.[1][7]

A DPIA describes the planned work, assesses necessity and proportionality, identifies risks to people and records measures to address them. Where high residual risk remains without sufficient mitigation, Article 36 requires prior consultation. Schedule that work early enough to change the workflow, rather than treating it as a final signature.[1]

Use the EU/EEA AI-agent DPIA reference to locate country lists and German Land publications. It keeps individual conditions, combined criteria and publication scope visible, including its disclosed Bulgarian-language substitution and Thuringian source gap. Bring the complete relevant clause back to the assessment.

The DPO test includes public authorities and bodies, and specified core activities involving large-scale regular systematic monitoring or large-scale special-category or criminal-offence data. National law can require appointment in additional cases. Record the applicable assessment and obtain qualified advice on uncertain points.[1]

Whether or not a formal appointment is required, assign someone to coordinate privacy work. Give that person access to supplier changes, incident reports and requests to expand the agent’s job. Naming a responsible contact without a route to those decisions leaves the operational work uncovered.

What happens when the agent exposes personal data?

Stop further exposure, preserve the facts and activate the incident process. A personal data breach may involve loss, alteration, unauthorised access or disclosure, rather than only a malicious attack. Articles 33 and 34 distinguish regulator notification from communication to affected individuals.[1][11]

The controller must notify the competent authority without undue delay and, where feasible, within 72 hours after awareness, unless risk to people is unlikely. A processor must notify its controller without undue delay. High risk can require communication to affected people without undue delay, subject to Article 34’s exceptions.[1][11]

Say an agent attaches one customer’s correspondence to another customer’s response. Stop similar sends while the cause is investigated. Record the information involved, recipients, time of awareness, likely consequences and containment. Ask whether retained copies or other sends are affected. Do not erase the only evidence while trying to remove the disclosure.

The response plan should name the decision-maker, supplier contacts, technical stop procedure and route to qualified advice. Document the breach, its effects and remedial action even where the notification threshold is not met. Record the reasoning rather than leaving an empty notification field.[1][11]

Reopening the workflow needs its own decision. Check that the failed boundary now works, whether permissions or instructions must change, and whether the processing description, DPIA or other records need updating. A successful restart is not the same as resolving the exposure.

What should be ready before the first live run?

Use a launch record that joins the legal decisions to tested controls. The checklist below is a practical way to organise the duties discussed here; it is not a regulator’s certification or a substitute for deciding the particular processing.

Download the European AI-agent deployer checklist (PDF). Assign an owner and evidence location to each line before marking it complete.

Launch questionEvidence to attachPerson who must resolve it
What job and information are approved?Processing description and purposeBusiness owner
What makes the use lawful?Basis and any additional data conditionsPrivacy decision-maker
Who handles the data?Roles, processor terms and recipientsSupplier owner
What have people been told?Relevant notices and contact routeCustomer or staff process owner
What happens to copies?Retention, rights and deletion testSystem and privacy owners
Can the agent cross a boundary?Access, approval and failure test resultsWorkflow operator
What happens if something goes wrong?Stop procedure and incident contactsIncident lead
What requires a new review?Change triggers and review ownerBusiness owner

Walk through one ordinary task and one exception with the people who will operate them. Include an unexpected recipient, a failed connection and an instruction to expand the job. These exercises are recommendations for finding gaps; they are not a measured success rate or a prescribed legal test suite.

Keep unresolved issues visible. Narrowing access, removing an unreviewed destination or retaining human execution may allow a smaller workflow to proceed once its own requirements are met. It should not quietly authorise the broader version that remains unresolved.

How do you keep the GDPR decisions current?

Review the processing when its purpose, data, recipients, retention, autonomy or consequences change. Keep the approved description, supplier arrangement and test evidence together so the operator can identify a request that falls outside it.[1]

A practical review cadence is to check the applicable guidance and list every six months, and at each material change. This is a suggested operating routine, not a GDPR deadline. Name the person who receives supplier notices and decides whether a proposed change needs privacy review.

For example, adding a tool that only reads one approved record differs from giving it permission to send that record to a new recipient. Review the additional operation before enabling it, and update notices, terms, records and the DPIA where needed. Retest the changed boundary with the person who operates it.

The UK has its own regime. The ICO’s current AI guidance carries an under-review notice following the Data (Use and Access) Act; use the applicable local rules for a UK operation.[6] The EU AI Act also needs its own review through the linked deployer guide.

Bring unresolved classifications, lawful bases and transfer questions to a qualified data-protection professional before the affected processing starts. The aim is a decision about the actual job, with changes made while they are still possible.

Frequently asked questions

Is my business the controller if an AI agent acts autonomously? Normally yes. The EDPB's Guidelines 07/2020 state that controller status follows an actor's actual activities rather than the formal designation in a contract. A business that decides why the agent runs and what data it may reach is determining the purposes and means of that processing. Autonomy in execution does not move that decision to the vendor.[2]

Do we need a DPIA before launching an AI agent? It depends on the processing, not on the word 'agent'. Article 35(1) requires a DPIA before processing that is likely to result in a high risk to rights and freedoms, and specifically mentions new technologies. Article 35(3) names systematic extensive automated evaluation underlying legally or similarly significantly affecting decisions, large-scale special-category or criminal-offence processing, and systematic large-scale monitoring of publicly accessible areas. Individual EU supervisory authorities also publish their own mandatory-DPIA lists, which should be checked directly.[1]

Does the GDPR small-business exemption cover records of processing? Rarely, once an agent is running. Article 30(5) exempts organisations with fewer than 250 employees, but the exemption fails if any one of three conditions applies: the processing is likely to result in a risk to rights and freedoms, the processing is not occasional, or it includes Article 9 special-category or Article 10 criminal-offence data. An agent that runs on a schedule is difficult to describe as occasional processing.[1]

Can we run an AI agent with customer data outside the EU? For transfers outside the EEA, assess Chapter 5 and the applicable transfer mechanism, safeguards or limited derogation, alongside the other GDPR duties. The practical difficulty with agents is knowing that a transfer happened. An agent calling a summarisation or translation service may move personal data across a border without that call appearing in any transfer record.[1]

Is this guide legal advice? No. This guide maps agent actions to the GDPR articles they commonly engage, and separates official text from guidance and interpretation. It cannot classify a specific business, resolve a specific processing operation, or replace advice from a qualified data protection professional.[7]

Notes

Sources accessed 15 October 2026. Legal texts, regulator guidance and the author’s control proposal are distinguished in the article. Practical examples and launch tests are recommendations, not enforcement cases.

  1. GDPR — official Publications Office consolidation. EU Publications Office; Regulation 2016/679; consolidation 4 May 2016.
  2. EDPB — Guidelines 07/2020. EDPB; adopted 7 July 2021; v2.1 minor corrections 20 September 2022.
  3. AEPD — Agentic artificial intelligence. AEPD; v1.1 February 2026.
  4. Navaie — Engineering GDPR compliance in the age of agentic AI. IAPP / Keivan Navaie; 8 October 2025.
  5. Eurostat — isoc_eb_ai. Eurostat; 2025 data; updated 15 June 2026.
  6. ICO — AI and data protection. ICO; page updated 15 March 2023; current review notice.
  7. EDPB — Be compliant. EDPB; living guide.
  8. EDPB — Process personal data lawfully. EDPB; living guide.
  9. EDPB — Respect individuals’ rights. EDPB; living guide.
  10. EDPB — International transfers. EDPB; living guide.
  11. EDPB — Data breaches. EDPB; living guide.