# DPIA vs FRIA: the two impact assessments for AI systems in Luxembourg

> A GDPR DPIA covers personal data risks; the AI Act FRIA covers all fundamental rights and applies to some high-risk deployers from 2 December 2027.

Published: 2026-10-22 · Dataplaz editorial team · https://dataplaz.lu/en/dpia-fria-ai-systems-luxembourg/

## Key takeaways
- A data protection impact assessment (DPIA) under Article 35 GDPR is required before any processing of personal data likely to result in a high risk, whether or not AI is involved.
- A fundamental rights impact assessment (FRIA) under Article 27 of the AI Act applies only to certain deployers of high-risk AI systems, before first use, from 2 December 2027 for systems listed in Annex III.
- The FRIA applies to public bodies, private entities providing public services, and deployers of AI for credit scoring of natural persons or for life and health insurance risk assessment and pricing.
- The CNPD's list of processing operations that require a DPIA was adopted by deliberation No. 34/2019 of 6 March 2019.
- Under draft law No. 8476, FRIA results from banks and insurers would be notified to the CSSF or the Commissariat aux assurances, and from most other deployers to the CNPD.

Luxembourg organisations that use AI on personal data can face two different impact assessments. The data protection impact assessment (DPIA) under Article 35 of the GDPR applies today to any high-risk processing of personal data, with or without AI. The fundamental rights impact assessment (FRIA) under Article 27 of the AI Act applies only to certain deployers of high-risk AI systems, such as banks scoring consumers' creditworthiness, and starts on 2 December 2027 for systems listed in Annex III.

## Two assessments, two questions

Consider a Luxembourg bank that buys a credit scoring model for consumer loans. Who in the bank owns the DPIA, and who will own the FRIA? If the answer is "the DPO, probably both", the bank has a gap waiting to happen.

The two documents ask different questions. The DPIA asks: what can go wrong for people because we process their personal data this way? The FRIA asks: what can go wrong for people's fundamental rights, including non-discrimination, access to services and an effective remedy, because we use this AI system in this process? The first is about data. The second is about decisions.

**A DPIA protects data subjects. A FRIA protects the people a system decides about.** Often these are the same people. Not always.

## Side by side

| | DPIA (GDPR) | FRIA (AI Act) |
|---|---|---|
| Legal basis | [Article 35 GDPR](https://gdpr-info.eu/art-35-gdpr/) | [Article 27 AI Act](https://eur-lex.europa.eu/eli/reg/2024/1689/oj) |
| Who | Any controller of personal data | Public bodies, private entities providing public services, deployers of credit scoring and life and health insurance pricing AI |
| Trigger | Processing likely to result in a high risk | First use of a high-risk AI system listed in Annex III (except point 2, critical infrastructure) |
| Applies | Already applies | From 2 December 2027 for Annex III systems |
| Scope | Rights and freedoms affected by the processing of personal data | Fundamental rights of the persons and groups likely to be affected |
| Filing | None by default; prior consultation of the CNPD in the cases set out in [Article 36](https://gdpr-info.eu/art-36-gdpr/) | Results notified to the market surveillance authority, using a template from the AI Office |

## When a DPIA is required

Article 35(3) GDPR names three cases where a DPIA is always needed: systematic and extensive evaluation of personal aspects by automated means, including profiling, that leads to decisions with legal or similarly significant effects; large-scale processing of special categories of data or criminal records; and large-scale systematic monitoring of a publicly accessible area.

In Luxembourg, the CNPD adds its own list. Its [deliberation No. 34/2019 of 6 March 2019](https://cnpd.public.lu/dam-assets/fr/actualites/national/2019/Delib-34-2019-du-6-mars-2019.pdf) names, among others, the combination or matching of data collected for different purposes when it produces legal or similarly significant effects, regular and systematic monitoring of employees' activities, and systematic tracking of people's location. The list is not exhaustive: the CNPD refers to the nine criteria of the European guidelines for anything else that is likely to be high-risk.

AI projects hit these triggers often. A CV screening tool, an employee productivity dashboard or a model that combines customer data from several sources can each require a DPIA today, years before any FRIA.

The minimum content is set by Article 35(7): a description of the processing and its purposes, an assessment of necessity and proportionality, an assessment of risks, and the measures to address them. The controller must seek the advice of its data protection officer.

## When a FRIA is required

The FRIA has a narrow scope. Under Article 27(1), it applies before first use of a high-risk AI system listed in Annex III, but only for two groups of deployers:

- bodies governed by public law and private entities providing public services;
- any deployer of AI used to evaluate the creditworthiness of natural persons or establish their credit score (Annex III point 5(b), which excludes fraud detection), or for risk assessment and pricing of natural persons in life and health insurance (point 5(c)).

For Luxembourg, the second group is the one that matters. It catches banks, consumer lenders and insurers.

The assessment must describe the deployer's process, the period and frequency of use, the categories of people and groups likely to be affected, the specific risks of harm, the human oversight measures, and what happens when risks materialise, including internal governance and complaint mechanisms. Deployers may rely on an earlier FRIA or one carried out by the provider in similar cases, but must update it when an element changes.

The results go to the market surveillance authority. Under Article 7 of [draft law No. 8476](https://wdocs-pub.chd.lu/docs/exped/0151/094/302942.pdf), filed on 23 December 2024, that would be the CSSF for entities it supervises, the Commissariat aux assurances for insurers and the CNPD as the default authority. The bill was not yet law at the time of writing, so these designations can still change.

## Where they overlap

The good news: the AI Act does not ask for the work to be done twice. Article 27(4) says that where an obligation is already met through a DPIA, the FRIA complements that DPIA. Article 26(9) adds that deployers must use the information the provider gives them under Article 13 to carry out their DPIA.

In practice, the system description, the categories of people affected and many risk mitigation measures can be written once. What the FRIA adds is the view beyond data: discrimination against groups, the effect of a wrong decision on access to credit or insurance, how a customer can contest the outcome, and who in the business can stop the system.

That last point matters. A DPIA is usually owned by the privacy function. A FRIA describes how a business process uses an AI system, so the business owner of that process has to sign it. **One system, one named owner.** Not the DPO, not IT, not a working group.

## What to do now

1. **List your AI use cases** and mark those that process personal data and meet a CNPD or Article 35(3) trigger. These need a DPIA today.
2. **Identify FRIA candidates**: credit scoring of individuals, life and health insurance pricing, and any high-risk AI used by a public body or in a public service. Plan for 2 December 2027.
3. **Ask your providers** for the Article 13 instructions for use and any assessment they have done. You will need both for the DPIA and the FRIA.
4. **Build one template** that holds the DPIA and adds a FRIA section on affected groups, human oversight and complaints, with a named business owner per system. If you want a second opinion on that structure, [an independent review of your AI governance set-up](https://harro.lu/work-with-me) is one way to test it.
5. **Follow draft law No. 8476** on chd.lu to see which authority will receive your FRIA notification.

## Questions & answers
**Is a FRIA the same as a DPIA?**
No. A DPIA under the GDPR looks at risks to people from processing their personal data. A FRIA under the AI Act looks at risks to all fundamental rights from using a high-risk AI system, and it is notified to the market surveillance authority.

**Can one document serve as both the DPIA and the FRIA?**
Partly. Article 27(4) of the AI Act says that where a DPIA already covers some of the FRIA obligations, the FRIA complements it. The DPIA can be reused, but the fundamental rights questions it does not answer still have to be covered.

**When does the FRIA obligation start?**
It applies with the other high-risk obligations: from 2 December 2027 for stand-alone systems listed in Annex III, as set by the Digital Omnibus, Regulation (EU) 2026/1744.

**Does a Luxembourg bank need a FRIA for its fraud detection model?**
Not on the basis of Annex III point 5(b), which excludes AI used to detect financial fraud. A DPIA may still be required under the GDPR if the processing is likely to result in a high risk.

## Sources
1. [Regulation (EU) 2024/1689 (AI Act)](https://eur-lex.europa.eu/eli/reg/2024/1689/oj) · EUR-Lex
2. [Regulation (EU) 2026/1744 (Digital Omnibus on AI)](https://eur-lex.europa.eu/eli/reg/2026/1744/oj) · EUR-Lex
3. [Délibération n° 34/2019 du 6 mars 2019 (liste des traitements soumis à une AIPD)](https://cnpd.public.lu/dam-assets/fr/actualites/national/2019/Delib-34-2019-du-6-mars-2019.pdf) · CNPD
4. [Projet de loi n° 8476 – document de dépôt (23 December 2024)](https://wdocs-pub.chd.lu/docs/exped/0151/094/302942.pdf) · Chambre des Députés
5. [Article 27: Fundamental rights impact assessment for high-risk AI systems](https://artificialintelligenceact.eu/article/27/) · artificialintelligenceact.eu
6. [Article 26: Obligations of deployers of high-risk AI systems](https://artificialintelligenceact.eu/article/26/) · artificialintelligenceact.eu
7. [Annex III: High-risk AI systems referred to in Article 6(2)](https://artificialintelligenceact.eu/annex/3/) · artificialintelligenceact.eu
8. [Article 35 GDPR: Data protection impact assessment](https://gdpr-info.eu/art-35-gdpr/) · gdpr-info.eu
9. [Article 36 GDPR: Prior consultation](https://gdpr-info.eu/art-36-gdpr/) · gdpr-info.eu
