legal247

Fundamental rights impact assessment: who must carry one out under the AI Act?

In brief

Under Article 27 of the AI Act, a fundamental rights impact assessment (FRIA) must be carried out before the first use of a high-risk system by bodies governed by public law, private entities providing public services, and deployers using AI for creditworthiness assessment or for risk assessment and pricing in life and health insurance. The obligation applies in the EU from 2 December 2027.

On 16 November 2026 the European Commission’s AI Office published a draft template for the assessment of the impact on fundamental rights which certain deployers of high-risk AI systems must carry out under AI Act Article 27. The template has been prepared under AI Act Article 27(5) with input from the EU Agency for Fundamental Rights (FRA), and it is to be used when the results are notified to the market surveillance authority. The Commission invites feedback until 15 January 2027, and organisations wishing to test the template may apply by 4 December 2026. This article considers which private businesses are covered, when the assessment must be carried out, how it relates to the data protection impact assessment (DPIA), and what a business should do now.

What is a fundamental rights impact assessment?

A fundamental rights impact assessment, commonly abbreviated FRIA, is a documented assessment of how the use of a specific high-risk system may affect the rights of the persons concerned. The obligation rests on the deployer, meaning the person using the system under its authority, as defined in AI Act Article 3(4), and not on the provider that developed it. Under AI Act Article 27(1), the assessment must describe the processes in which the system will be used, the period and frequency of use, the categories of persons and groups likely to be affected, and the specific risks of harm to them. It must also describe how human oversight is implemented in accordance with the instructions for use, and the measures to be taken if the risks materialise, including internal governance and complaint mechanisms.

The assessment complements the provider’s risk management system under AI Act Article 9. The Commission’s draft observes that the provider’s risk management covers only risks which may reasonably be mitigated or eliminated through development, design or the provision of technical information. Risks arising from the deployer’s particular context of use are therefore to be captured in the impact assessment.

Which private businesses must carry out a fundamental rights impact assessment?

The obligation applies to four groups of deployers only, and only in respect of high-risk systems under AI Act Article 6(2), that is the systems listed in Annex III. Systems used for critical infrastructure under Annex III, point 2, are expressly excluded. Among private businesses, the financial sector and providers of services of a public nature are those principally affected.

Deployer Systems Legal basis
Bodies governed by public law High-risk systems in Annex III, except point 2 AI Act Article 27(1)
Private entities providing public services High-risk systems in Annex III, except point 2 AI Act Article 27(1)
Any deployer assessing creditworthiness Creditworthiness and credit scores of natural persons, excluding fraud detection Annex III, point 5(b)
Any deployer pricing life and health insurance Risk assessment and pricing for natural persons Annex III, point 5(c)

For banks, finance companies and insurers, the obligation depends on the purpose of use rather than on the type of business using the system. A system that estimates the creditworthiness of an individual applying for a loan is covered, whereas a system used solely to detect financial fraud falls outside point 5(b). The same applies to other businesses extending credit to consumers, for example through instalment payments in online retail.

The concept of private entities providing public services is not defined in the Regulation. Recital 96 mentions education, healthcare, social services, housing and the administration of justice as examples, and the Commission’s draft takes the view that the list is not exhaustive. Private schools, private healthcare providers under public contracts and private operators in child welfare and social care should therefore consider whether they are covered when deploying a high-risk system. In our assessment, the wording and purpose of the provision indicate that decisive weight should be given to whether the service is of a public nature and important to individuals, rather than to how the provider is organised.

When must the assessment be carried out and updated?

The assessment must be carried out before the high-risk system is first deployed, pursuant to AI Act Article 27(1) and (2). According to the Commission’s draft, it must be carried out for each system and cover all the purposes for which the system is used. The deployer must update the assessment if any of the elements in paragraph 1 has changed or is no longer up to date. The draft gives changed processes, new groups of affected persons and a new version of the system or its underlying model as examples.

In similar cases, the deployer may rely on previously conducted impact assessments, including assessments carried out by the provider, under AI Act Article 27(2). The Commission states in the draft that the provider is not obliged to share its assessment with the deployer, although sharing may constitute good practice. For businesses procuring a credit scoring system or a pricing tool, access to the provider’s assessment is therefore a matter for the contract.

How does the assessment relate to the DPIA?

The fundamental rights impact assessment is additional to the data protection impact assessment under GDPR Article 35, although the two can largely build on each other. Under AI Act Article 27(4), as amended by Regulation (EU) 2026/1744, the deployer may cross-refer to relevant sections of the DPIA, or include relevant parts of it, where obligations are already met through the DPIA. Under the new wording of Article 27(5), the template is to allow such cross-references.

“the deployer may, when conducting the fundamental rights impact assessment referred to in paragraph 1 of this Article, include cross-references to the relevant sections of that data protection impact assessment”

The draft template reflects this in section 4.1, where the assessment of the right to the protection of personal data will typically correspond to the risk assessment and measures in the DPIA. The impact assessment nevertheless goes further, because section 4.2 requires a separate assessment of other rights under the EU Charter of Fundamental Rights, including equality and non-discrimination, the right to an effective remedy and the rights of the child, the elderly and persons with disabilities. A DPIA prepared in line with the article on DPIAs for AI tools therefore covers only part of the assessment. The wider regulatory framework is discussed on the artificial intelligence topic page.

What does the Commission’s draft template contain?

The draft follows the structure of AI Act Article 27(1) and has seven sections. The first section covers general information, including the type of deployer, whether this is the first deployment, and whether the deployer relies on an earlier assessment. It is followed by the context of use, affected groups, risks of harm to fundamental rights, human oversight, measures in case risks materialise, and an overall conclusion. Likelihood and severity are to be rated on a scale of low, medium and high, and preventive measures are not required for risks rated low.

The section on human oversight links the assessment to AI Act Article 26(2). The deployer must describe who exercises oversight, what authority they have to override or disregard the system’s output, and what training they have received, including measures to support AI literacy under AI Act Article 4. Any departure from the provider’s instructions for use must be explained. The section on measures must describe monitoring, the duty to inform the provider under AI Act Article 26(5) and how complaints from affected persons are handled.

Once the assessment has been carried out, the deployer must notify the market surveillance authority of the results and attach the completed template, under AI Act Article 27(3). According to the draft, the exemption for systems authorised by the market surveillance authority under AI Act Article 46(1) concerns only the notification and not the assessment itself. A summary is published in the EU database only for public authorities and persons acting on their behalf, under AI Act Article 49(3), so private banks and insurers are not required to publish their assessments.

When does the obligation apply, and what does it mean in Norway?

In the EU, AI Act Article 27 applies from 2 December 2027 to high-risk systems listed in Annex III. The date follows from AI Act Article 113 as amended by Regulation (EU) 2026/1744, which postponed the application of Sections 1 to 3 of Chapter III from 2 August 2026. The draft template has not been adopted, and the Commission states that the final template will be made available as an automated tool.

In Norway, the obligation does not yet apply. The AI Act has not been incorporated into the EEA Agreement, and EEA-Lex lists it as under scrutiny. The Norwegian Government has put the AI Omnibus amendments out to consultation with a deadline of 11 December 2026. For high-risk systems used by financial institutions in direct connection with the provision of financial services, the starting point under AI Act Article 74(6) is that the financial supervisory authority acts as market surveillance authority, while AI Act Article 74(7) allows another authority to be designated. Which Norwegian authority will receive notifications from other deployers has not been laid down in Norwegian law. Norwegian banks and insurers procuring credit scoring or pricing systems now will nonetheless normally be using them once the rules take effect, and the assessment must then be in place before the system is deployed.

What should the business do?

Work on the impact assessment should be tied to the procurement and to the DPIA, because much of the information comes from the provider. The following measures should be prioritised.

  1. Identify the AI systems the business uses or plans to use for creditworthiness assessment, credit scoring or risk assessment and pricing in life and health insurance, and whether the business provides services of a public nature.
  2. Distinguish systems used solely for fraud detection from systems that also influence credit decisions, and document that assessment.
  3. Require the provider to supply the instructions for use and information under AI Act Article 13, and agree access to the provider’s own impact assessment, as discussed in the article on AI vendor terms for high-risk use.
  4. Structure the DPIA so that its sections on risks and measures can be used directly in section 4.1 of the template, and add the assessment of discrimination and access to complaints.
  5. Designate the persons responsible for human oversight, and document their authority and training in line with the AI literacy requirement.
  6. Consider submitting feedback to the Commission by 15 January 2027, or applying to test the template by 4 December 2026, particularly where the business has experience with DPIAs for credit scoring or insurance pricing.

Questions and answers

Must a bank carry out a fundamental rights impact assessment for an AI system that detects fraud?

Not on account of fraud detection alone. Annex III, point 5(b), excludes AI systems used for the purpose of detecting financial fraud from the creditworthiness category. If the same system is also used to evaluate creditworthiness or establish a credit score, that part of its use is covered.

May the business rely on the provider's own impact assessment?

Yes, in similar cases. Under Article 27(2), the deployer may rely on previously conducted assessments or on existing impact assessments carried out by the provider. According to the Commission's draft, the provider is not obliged to share its assessment, so access to it should be agreed in the contract.

Must a private bank publish a summary of the assessment?

No. Under the draft, only public authorities, Union bodies and persons acting on their behalf register the use under Article 49(3) and have a summary published in the EU database. The bank must nevertheless notify the market surveillance authority of the results.

Next legal review: 31 March 2027