Skip to main content
Praxikon

The draft FRIA template: what the AI Office asks of deployers

··14 min read

A municipality wants to use an AI system for applications for special social assistance. The system advises the caseworker on whether the applicant is likely to qualify. If it puts the system into use on or after 2 December 2027, it first has to carry out a fundamental rights impact assessment: the FRIA under Article 27. If the system is already running before then, Article 111(2) applies (see below). On 16 November 2026 the European Commission published a draft of the template that the AI Office has to develop for this under Article 27(5), together with a draft explanatory notice.

What Article 27 itself requires is covered in our explanation of the FRIA and Article 27. This piece is about the draft and what it gives you as a deployer. The municipality, the bank and the insurer are fictitious.

In short

The draft has not been adopted by the Commission: it reflects the preliminary views of the Commission services. The template follows Article 27(1), has you rate the likelihood and severity of each risk as low, medium or high and leaves room for your DPIA. Feedback is open until 15 January 2027, and deployers can sign up to test the draft until 4 December 2026. The obligation applies from 2 December 2027; for systems already on the market or in use before then, Article 111(2) applies.

What was published on 16 November

The document has two parts: an explanatory notice and a template with seven sections plus a summary. The AI Office drafted it with detailed input from the EU Agency for Fundamental Rights (FRA). Its first page says it has not been adopted or endorsed by the Commission and that it contains the "preliminary views of the Commission services". According to the news item, once the template is final, deployers use it for their FRIA and notify the results with it to the market surveillance authority (Article 27(3)). After adoption, it will also become an automated tool on the AI Act Single Information Platform.

Who has to carry out a FRIA

The notice names three groups of deployers: bodies governed by public law and private entities providing public services, when they use a high-risk system from Annex III; deployers of systems that evaluate the creditworthiness of natural persons or establish their credit score, except systems for detecting financial fraud (point 5(b)); and deployers of systems for risk assessment and pricing in life and health insurance (point 5(c)). Systems for critical infrastructure (point 2) are explicitly excluded.

The AI Act does not define "body governed by public law". According to the notice, the case law of the Court of Justice on "public authority" and "body governed by public law" in other fields of EU law may provide inspiration. The concept may then cover State organs and regional and local authorities, irrespective of whether they have separate legal personality, as well as other entities established under public law to perform public tasks, entities entrusted by law with public authority, and Union institutions and bodies. A municipality is a local authority and so falls within it on this reading; the interpretation of the Court and Article 2 take precedence.

For private entities the notice relies on Recital 96, which mentions education, healthcare, social services, housing and the administration of justice. The notice treats these as examples: a private entity providing a similar service of a public nature should also carry out a FRIA.

When to carry it out and update it

The FRIA has to be done before the first deployment. According to the notice, you carry out one per AI system, covering all its purposes of use, and update it (Article 27(2)) when a relevant factor changes, such as the processes concerned, the persons affected, the version of the system or its underlying model, or when the assessment is no longer up to date.

The draft itself gives no date. The news item mentions 2 December 2027, the date from which the Annex III high-risk rules apply under Regulation (EU) 2026/1744. For a system placed on the market or put into service before that date, Article 111(2) applies: the high-risk obligations, including the FRIA, only apply once the system undergoes a significant change in its design, and systems intended to be used by public authorities must comply by 2 August 2030 at the latest. The prohibitions in Article 5 apply regardless. Whether a new version of the underlying model counts as such a significant change is not addressed in the draft.

What the provider contributes

The provider runs a risk management system (Article 9), but paragraph 3 limits it to risks "which may be reasonably mitigated or eliminated through the development or design" of the system or the provision of adequate technical information. Risks that arise from the use and the individual deployment context can therefore be captured in the FRIA, the notice says. You should rely on the instructions for use (Article 13(2) and (3)), which according to the notice should include the information from that risk management system; Article 27(1)(d) requires you to take that information into account.

In similar cases you may build on a previous FRIA or on an existing impact assessment by the provider (Article 27(2)); for the provider's assessment, the template says this is also possible at your first use. The provider is not obliged to share its assessment, although the notice calls sharing a possible good practice. In our reading, that is a reason to ask about it during procurement.

The DPIA as a building block

According to the notice, the FRIA complements the DPIA. Where a DPIA for the processing around the system already covers part of the FRIA, you may cross-refer to the relevant parts or include them (Article 27(4) and (5)). The template works this out in Table 4.1, on the right to the protection of personal data. You indicate whether there is a DPIA under Article 35 GDPR, Article 27 of Directive (EU) 2016/680 or Article 39 of Regulation (EU) 2018/1725 (EUDPR). The risk description will typically match the assessment under Article 35(7)(c) GDPR, and the measures point (d). Where the DPIA rates likelihood and severity, you restate that on the template's scale. For more on how the two differ, see DPIA vs FRIA: a practical comparison.

The template, section by section

Section 1, general information. Type of organisation, responsible person, the system's identification number in the EU database (Article 49(1)) and, where applicable, your own registration number (Article 49(3)). You tick the Annex III area and say whether this is a first deployment or an update.

Section 2, context of use (Article 27(1)(a) and (b)). The intended purpose under the instructions for use, the processes in which the system works, how its output is used, and the period and frequency of use.

Section 3, who is affected (point (c)). A checklist that includes children, older people, persons with disabilities, persons in a vulnerable social or economic situation and third country nationals. "The general public" does not replace those groups.

Section 4, risks of harm (point (d)). After Table 4.1 comes 4.2 for other fundamental rights. For each risk, a concrete way the system could harm people, you describe it, tick the affected rights from a Charter-based list and name the groups. Risk is the combination of the probability of harm and its severity (Article 3, point (2)). You rate both as low, medium or high. Low means unlikely under foreseeable conditions, medium plausible under some conditions, high highly probable or already observed in comparable deployments. According to the template, preventive and mitigating measures are not required for low risk; the measures for when risks materialise (Section 6) are still needed. You then rate the risk that remains after the measures.

Section 5, human oversight (point (e)). In 5.1 you describe which measures from the instructions for use you have implemented, who carries out oversight, what authority they have to override, disregard or reverse the output, and where you deviate from the instructions. In 5.2 you describe how the people assigned to oversight get the necessary competence, training, authority and support (Article 26(2)), including measures for AI literacy (Article 4).

Section 6, if risks materialise (point (f)). In 6.1 you describe monitoring and your immediate response, such as suspending use, with the reporting duties to the provider under Article 26(5). Note that where there is a risk within the meaning of Article 79(1), Article 26(5) requires you to suspend use and also inform the market surveillance authority. In 6.2 you describe complaint handling, internal governance and any redress mechanisms.

Section 7 and the summary. Section 7 is a short synthesis per right assessed. The summary is only filled in by deployers that register their use under Article 49(3): public authorities, Union institutions and bodies, and persons acting on their behalf. It is published in the public EU database (Article 49(3), Article 71 and Annex VIII, Section C, point 4). The summary does not apply to systems registered under Article 49(4) (law enforcement, migration, asylum and border control).

Notifying the market surveillance authority

Once the template is completed or updated, you notify the results to the market surveillance authority of your Member State, attaching the filled-in template. Where that authority has exceptionally authorised the system, for example for reasons of public security (Article 46(1)), you may be exempt from notifying; according to the notice, this covers the notification only, not the FRIA itself. Under the conditions of Article 77, authorities protecting fundamental rights may request access to the FRIA through the market surveillance authority. Which authority will receive notifications in the Netherlands has not yet been settled.

The municipality, the bank and the insurer

The municipality's system helps decide whether applicants qualify for a public benefit; in our reading it falls under Annex III, point 5(a). In Section 3 the municipality lists persons in a vulnerable social or economic situation, persons with disabilities and older people. Table 4.1 cross-refers to the DPIA. Risk 1 in 4.2: the system advises negatively more often for applicants with fluctuating income, while that group is often entitled to assistance. Affected are the right to social security and social assistance, non-discrimination and good administration; likelihood medium, severity high. Measures: the caseworker always reviews the full application, a negative advice never leads to a refusal on its own, and there is a monthly sample check. In Section 6 it compares advice with decisions per group; it then publishes the summary in the EU database. If the system is already running before 2 December 2027, Article 111(2) applies: the FRIA is needed upon a significant change in its design, and by 2 August 2030 at the latest.

A bank that uses AI to assess the creditworthiness of natural persons also has to carry out a FRIA, even though it is a private party; the summary for the EU database does not apply to it. For financial institutions, the monitoring duty in Article 26(5) is deemed fulfilled by complying with the governance rules of financial services law (second subparagraph); the duties to inform and to suspend use in the first subparagraph still apply. In our reading the bank can refer to those rules in Section 6 for monitoring; the draft says nothing about it. An insurer using AI for risk assessment and pricing in relation to natural persons in life or health insurance falls under point 5(c), with older people and persons with disabilities as obvious groups for Section 3.

What is still open

  • Three references still read "[link to be added later]": where to submit an update, which authority can give advice, and how to notify.
  • Low, medium and high are described in terms of likelihood, while Table 4.1 also refers to those descriptions for severity; a description that fits severity is missing.
  • Table 4.1 drops measures only where both likelihood and severity are low; 4.2 simply says "low risk". How to treat a low likelihood with a high severity is open.

What to do now

  1. Give feedback on the draft until 15 January 2027 through the Commission's consultation.
  2. Sign up to test it by 4 December 2026 through the expression of interest form. You apply the template to one of your own systems, supported by the AI Office; the AI Office may have to select the cases.
  3. Get your DPIA in order, with a rating of likelihood and severity you can restate on the template's scale.
  4. Obtain the instructions for use from the provider, and ask whether it will share an impact assessment.
  5. Set up a team with knowledge of fundamental rights, the system and the work process, and decide who is accountable for the final assessment.

Which obligations apply to your use case is shown in the case check. A working template is available in our FRIA template; for public bodies there is the FRIA for municipalities and the public sector.

Frequently asked questions about the draft FRIA template

Sources

EUR-Lex: Regulation (EU) 2026/1744 (Digital Omnibus on AI) (8 July 2026, Official Journal 24 July 2026, accessed 22 November 2026)
EUR-Lex: Regulation (EU) 2016/679 (GDPR), Article 35 (27 April 2016, accessed 22 November 2026)

Newsletter

Every Tuesday, the AI Act week ahead in 5 minutes

A practical briefing on deadlines, new guidance and enforcement, so you know what matters this week. No spam and you can unsubscribe in one click.

Practical and short · No spam · One-click unsubscribe

Written by

Zahed Ashkara, freelance AI and privacy consultant and legal adviser

Zahed Ashkara

Freelance AI and privacy consultant and legal adviser · AI governance & safety

Zahed Ashkara is a freelance AI and privacy consultant and legal adviser for the GDPR and the EU AI Act. Organisations hire him as AI governance consultant, privacy counsel, external DPO or AI governance lead, have him carry out a DPIA, FRIA or AI Act classification, or book an in-company training. He focuses on AI governance and safety, including AI agents. He works exclusively as a freelancer, through Embed AI (trade name Praxikon). He is a Certified AI Compliance Officer (CAICO) and has been a member of the NEN standards committee Artificial Intelligence & Big Data since July 2026.

Follow on LinkedInView expert profile

Use these sources in your AI assistant too: Praxikon in Claude and other assistants

Real-World Stories

Discover how organizations apply this in practice

Related articles

Deployer obligations for high-risk AI: what Article 26 of the AI Act requires

An organisation that uses a high-risk AI system has twelve paragraphs of obligations of its own (Article 26 AI Act): use in accordance with the instructions, human oversight by competent people, sound input data, monitoring and notification, keeping logs, informing workers and affected persons, registration by public authorities, feeding the DPIA and cooperating with the authorities. What each paragraph requires, what the Digital Omnibus did and did not change, the rules for banks and insurers, the difference with the provider, and how the FRIA and the right to explanation connect, with a municipality, a bank and an employer as examples.

EU AI ActArticle 26Deployer

How the Commission supervises general-purpose AI models: Articles 90 to 94 of the AI Act

Articles 90 to 94 of the AI Act set out four steps towards providers of general-purpose AI models: a qualified alert from the scientific panel to the AI Office, followed by the Commission's request for documentation and information, an evaluation of the model and a request for measures, with procedural rights for the provider. What each paragraph says, how Articles 101 and 113 come in, and what a municipality, bank or university of applied sciences that builds such a model into its own system will notice and record.

EU AI ActArticle 90Article 91