Two organizations can buy the same high-risk AI system and end up with different legal duties. Under the EU AI Act Article 27 fundamental rights impact assessment (FRIA), a public body or a private entity providing public services must assess most Annex III systems before deploying them. A bank or insurer must do the assessment only for a narrow set of uses, but in those cases it applies no matter what kind of organization is using the system. This explainer covers who has to carry out a FRIA, the six things it must contain, what happens once it is done, and when the obligation starts.
Who has to do a FRIA?
There are two separate tests, and they work differently.
The first test is about who the deployer is. Before deploying a high-risk system referred to in Article 6(2), meaning the Annex III list, deployers that are bodies governed by public law, or are private entities providing public services must assess the system's impact on fundamental rights. High-risk AI systems intended to be used in the area listed in point 2 of Annex III are excluded from this obligation. Annex III reaches into much of what public bodies do. It includes systems used to evaluate eligibility for essential public assistance benefits and services, including healthcare services, tools that evaluate and classify emergency calls or dispatch first responders, and systems used to examine applications for asylum, visa or residence permits.
The second test is about what the system does. Any deployer of a system in points 5(b) and (c) of Annex III must also carry out a FRIA. Point 5(b) covers AI systems intended to be used to evaluate the creditworthiness of natural persons or establish their credit score, with the exception of AI systems used for the purpose of detecting financial fraud. Point 5(c) covers AI systems intended to be used for risk assessment and pricing in relation to natural persons in the case of life and health insurance.
A building permit works the same way. Some permits depend on who is building, such as a public authority. Others depend on what is being built, whoever builds it. A lender's credit-scoring model falls under point 5(b) and triggers a FRIA. The same lender's fraud-detection model falls outside it.
Six items every assessment must contain
Article 27(1) says deployers shall perform an assessment consisting of six parts:
- Processes: a description of the deployer's processes in which the system will be used in line with its intended purpose.
- Time and frequency: the period of time within which, and the frequency with which the system is intended to be used.
- Affected people: the categories of natural persons and groups likely to be affected by its use in the specific context.
- Risks of harm: the specific risks to those groups, taking into account the information given by the provider pursuant to Article 13.
- Human oversight: how human oversight measures will be implemented according to the instructions for use.
- Response plan: the measures to be taken in the case of the materialisation of those risks, including the arrangements for internal governance and complaint mechanisms.
The assessment must take into account the information given by the provider pursuant to Article 13 regarding the specific risks of harm. Procurement teams can raise this during contract talks.
What happens after the first assessment?
The obligation applies to the first use of the high-risk AI system. It does not end there. If the deployer considers during use that any of the six elements has changed or is no longer up to date, it must take the necessary steps to update the information. In practice, the assessment should be treated as a document you keep up to date.
You don't always have to start from scratch. In similar cases, a deployer may rely on previously conducted fundamental rights impact assessments or existing impact assessments carried out by provider. Once the assessment is done, the deployer must notify the market surveillance authority of its results, submitting the filled-out template as part of the notification. In the case referred to in Article 46(1), deployers may be exempt from that obligation to notify. The AI Office shall develop a template for a questionnaire, including through an automated tool, so that deployers can meet their obligations more simply.
Many deployers already run data protection impact assessments. Where a GDPR Article 35 assessment, or one under Article 27 of Directive (EU) 2016/680, already meets part of the obligation, the deployer may include cross-references to the relevant sections or copy relevant parts into the FRIA. The template must also allow for those cross-references.
The Service Desk notes that its plain-language summaries are not legal binding. For compliance decisions, rely on the article text itself.
When does the AI fundamental rights assessment become mandatory?
The headline application date is misleading here. The Regulation shall apply from 2 August 2026, but the amended Article 113 sets a later date for this part. Chapter III, Sections 1, 2, and 3, with the exception of Article 6(5), apply from 2 December 2027 as regards AI systems classified as high-risk pursuant to Article 6(2) and Annex III.
| Date | Event | Impact |
|---|---|---|
| 2 February 2025 | Chapters I and II apply | General provisions, prohibitions |
| 2 August 2025 | Chapter III Section 4, Chapters V, VII, XII apply | Notified bodies, general-purpose AI models, governance, penalties |
| 27 July 2026 | Articles 102 to 110 apply | Amendments to other EU laws |
| 2 August 2026 | General application date | Most of the Regulation applies |
| 2 December 2026 | Article 5(1a) and (1b) apply | Added prohibitions take effect |
| 2 December 2027 | Annex III high-risk rules apply | Includes Article 27, the FRIA |
| 2 August 2028 | Annex I high-risk rules apply | Product-safety high-risk systems |
Getting ready for the FRIA
The assessment asks deployers to write down processes, timeframes, affected groups, risks, oversight arrangements and remedies. Three questions decide how much work this involves for your organization:
- Are you a public body, or a private entity providing public services?
- Do any of your systems score credit, or assess risk and set prices for life or health insurance?
- Can your existing data protection impact assessments be reused?
The obligation applies to the first use of the high-risk AI system. It also has to be updated whenever any of the six elements changes.
Sources
- ai-act-service-desk.ec.europa.eu – "AI Act Service Desk - Article 27: Fundamental rights impact assessment for high-risk AI systems"
- ai-act-service-desk.ec.europa.eu – "AI Act Service Desk - Annex III: High-risk AI systems referred to in Article 6(2)"
- ai-act-service-desk.ec.europa.eu – "AI Act Service Desk - Article 113: Entry into force and application"