Regulation (EU) 2024/1689 · Artificial Intelligence Act ● Annex III: — days

EU AI Act vs GDPR: What Changes in Your Existing Compliance Programme

The AI Act is not a second GDPR. It shares the paperwork in some places and diverges sharply in others — this is the map of where your existing programme carries over and where it does not.

Renomira Team · 10 min read

Most EU mid-market companies already run a GDPR programme: a record of processing, DPIAs, a retention schedule, a named DPO, a breach process. The natural instinct is to ask whether the AI Act is “GDPR for AI”.

A padlock on a door — where data protection and AI regulation meet
Photo: User:Nino Barbieri / CC BY-SA 2.5 via Wikimedia Commons

It is not. It shares three artefacts with GDPR, diverges on five, and introduces two obligations GDPR has no analogue for. Knowing which is which is the difference between a two-quarter extension and an eighteen-month rebuild.

Where they genuinely overlap

Records of processing ↔ AI register. Article 30 records and an AI register are similar instruments with different subjects: GDPR documents processing activities, the AI Act documents AI systems. There is heavy overlap — an AI tool processing CVs appears in both — but the AI register carries fields GDPR never asked for: risk tier, classification rationale, provider/deployer role, oversight owner, review date.

DPIA ↔ FRIA. Both are ex-ante impact assessments of a system affecting people. The trigger conditions differ sharply:

DPIA (GDPR Art. 35) FRIA (AI Act Art. 27)
Trigger High risk to rights and freedoms — systematic monitoring, large-scale profiling, special-category data Public-body deployer, private entity providing public services, or deployer of Annex III 5(b) creditworthiness / 5(c) essential services
Subject Personal data processing Fundamental rights impact, more broadly
Data required? Only personal data Not necessarily — it can apply without personal data
Supervised by Data protection authority Market surveillance authority

The practical point: you may need both for the same system, and they are not substitutes. A credit-scoring deployer with a DPIA still owes a FRIA. An AI system with no personal data owes no DPIA but may owe a FRIA.

Automated decision-making. GDPR Article 22 restricts solely automated decisions with legal or similarly significant effects, and grants rights around them. AI Act Article 86 grants a right to a clear, meaningful explanation of the role the AI played in a decision producing legal or similarly significant effects. They are complementary, differently scoped, and both need a process — usually the same one.

Where they diverge

1. Scope without personal data. The AI Act applies to an AI system whether or not it processes personal data. A model ranking machine parts, detecting manufacturing defects or optimising logistics carries AI Act obligations and no GDPR relevance at all. Your GDPR programme will not see these systems.

2. Product safety, not just data. Annex III point 1 covers biometric identification, and Article 6(1) pulls in AI as a safety component of products regulated under Union harmonisation legislation — medical devices, machinery, toys, lifts. That is conformity-assessment territory, with CE marking and a notified body, none of which GDPR has any concept of.

3. Transparency to the individual, not the data subject. Article 50 disclosure duties run to anyone interacting with the system, including people whose personal data you never touch. A public chatbot must say it is a machine even if it processes nothing.

4. AI literacy. Article 4 has no GDPR analogue. Your privacy-awareness training does not satisfy it, because the subject is model capability and limitation, not data handling.

5. Enforcement style. GDPR enforcement follows harm to individuals. AI Act enforcement includes market surveillance of the system itself: documentation, conformity, registration, monitoring. A perfectly privacy-respecting system can still be non-conformant.

Two things GDPR has no equivalent for

Post-market monitoring and incident reporting. Articles 72 and 73 require a monitoring plan per high-risk system and notification of serious incidents — generally within 15 days, 10 days for serious disruption of critical infrastructure, 2 days where there is death or serious harm to health and safety. GDPR’s 72-hour breach clock is a different instrument covering a different event; you may well run both from one incident.

Versioned classification. GDPR classifies processing once and revisits it on change. The AI Act classifies a system against a rule pack that moves — new guidance, a Commission interpretation, an amended annex. Re-classification is a recurring obligation, not a change-management exception.

Extend the programme you already have

The efficient path is augmentation, not parallel construction.

1. Extend the record of processing into a register. Add five columns: AI flag, risk tier, classification rationale, oversight owner, review date. Every existing record that names a tool is 80% of a register row.

2. Add a classification gate to your procurement workflow. “Does this tool rank, score or screen people? Does it talk to customers? Does it generate media?” Three questions at purchase prevent most of the exposure.

3. Reuse the DPIA template, add an Article 27 section. One assessment, two headers — with a documented reasoning block for the FRIA trigger.

4. Extend the awareness training. Keep the privacy module, add a literacy module: hallucination, prohibited practices, what the tiers mean, escalation. One course, two objectives, one attendance record.

5. Add the transparency register. Article 50 disclosure wording and placement per surface. Nothing in the GDPR programme covers this.

6. Put re-classification on the same quarterly rhythm as your RoPA review. One calendar, one owner.

What stays with the DPO, and what moves

The DPO retains everything about personal data: lawful basis, data-subject rights, breach, transfers. The AI Act brings a different owner for conformity: technical documentation, oversight, monitoring, incident reporting to market surveillance, and registration.

For a mid-market company, that is rarely a new hire. It is usually the CISO or Head of Compliance holding the conformity file, with the DPO sitting on the same review. What matters is that both roles exist and are named — “who is accountable for AI” is the question every questionnaire and every inspection opens with.

The order of operations

If you are starting from a functioning GDPR programme:

  1. Register — extend the RoPA; this is the foundation everything else keys off.
  2. Classification — tier each system with reasoning, screening Article 5 first.
  3. Remove prohibited uses — the only item with an absolute deadline behind it.
  4. Close the two GDPR-blind spots — systems with no personal data, and Article 50 transparency.
  5. Add literacy — cheapest evidence, earliest deadline.
  6. Build the conformity file — oversight, logs, monitoring, incidents, assessments.
  7. Schedule re-classification — quarterly, alongside the RoPA review.

Frequently asked questions

Does the EU AI Act replace GDPR for AI systems?

No. Both apply simultaneously and independently. An AI system processing personal data must satisfy GDPR for the data processing and the AI Act for the system itself — classification, documentation, oversight, transparency. Compliance with one is never a defence against the other.

Is a DPIA enough instead of a fundamental rights impact assessment?

No. A DPIA under GDPR Article 35 and a FRIA under AI Act Article 27 have different triggers, subjects and supervising authorities. Where both apply you need both, although a single assessment document can satisfy both with distinct sections and a documented reasoning for each trigger.

Do we need an AI register if we already have a record of processing?

In practice, yes — though you can build one from the other. A record of processing covers activities involving personal data; the AI register covers AI systems regardless of personal data, and adds risk tier, classification rationale, provider/deployer role and oversight owner. Extend your existing RoPA rather than starting over.

Who owns AI Act compliance — the DPO or the CISO?

Where personal data is involved the DPO retains ownership; conformity matters — technical documentation, oversight, monitoring, incident notification to market surveillance authorities and registration — normally sit with the CISO or Head of Compliance. Both roles must be named and cooperate on the same quarterly review.

Does GDPR awareness training satisfy AI literacy under Article 4?

No. AI literacy requires an understanding of what the systems do and do not do: hallucination, prohibited practices, synthetic-content marking, human oversight and the obligations attached to high-risk use. Privacy training covers data handling. Most companies deliver both as one course with two modules.

Next step

Put this guide to work on your own estate.

Run the free exposure scan to see which of these articles actually apply to your company — 12 questions, four minutes, no card. Confirm your email and the full report unlocks.

  • Free forever plan
  • No credit card
  • EU-hosted · GDPR-native
Run a free scan Sign up free

Build a live AI register in minutes — upgrade only when you need the audit pack.

EU AI Act vs GDPRAI Act and GDPR overlapGDPR DPIA vs AI Act FRIAdoes the AI Act replace GDPRAI Act compliance for DPOsGDPR Article 22 vs AI Actextend GDPR programme to AI Act

Keep reading