The AI Office’s Draft FRIA Template: What Deployers of High-Risk AI Should Know

On 16 November 2026, the European Commission published for consultation a draft template for the fundamental rights impact assessment (FRIA) required by Article 27 of the AI Act. It comes with a draft explanatory notice. This is the first official model of the FRIA. Until now, deployers have had to rely on the text of Article 27 and on methodologies developed by academics, authorities and civil society.
We looked at the FRIA from a methodological perspective in Fundamental rights impact assessment and AI: a new methodology for EU-funded research. Here we focus on the official model and on what it means for organisations that use high-risk AI systems.
The template was developed by the AI Office under Article 27(5), with detailed input from the EU Agency for Fundamental Rights (FRA). Deployers will use it both to carry out the assessment and to notify the results to the market surveillance authority. Once adopted, it will become an automated tool on the AI Act Single Information Platform. The document is still a draft: it has not been adopted by the Commission and reflects the preliminary views of its services.
Who must carry out a FRIA
The FRIA is an obligation of deployers, not providers. If you are unsure which role you have, see Provider or Deployer? How to Identify Your Role under the AI Act. Not every deployer of a high-risk system is covered. The notice lists three groups:
- Bodies governed by public law and private entities providing public services that use a high-risk system in one of the Annex III areas, except critical infrastructure (point 2). Deployers in the critical infrastructure area are expressly exempted.
- Deployers of systems that evaluate creditworthiness or establish a credit score. Systems intended for detecting financial fraud are excluded.
- Deployers of systems used for risk assessment and pricing in life and health insurance.
The AI Act does not define “bodies governed by public law”. The notice suggests a broad reading, drawing on the case law of the Court of Justice in other areas of EU law. The concept would cover:
- State organs and regional and local authorities, including legislative, executive, administrative, law enforcement and judicial authorities;
- other entities established under public law to perform public tasks;
- entities entrusted by law with public authority;
- EU institutions, bodies, offices and agencies.
For private entities, the notice relies on Recital 96 of the AI Act, which describes public services as services important for individuals that are of a public nature. The recital gives examples: education, healthcare, social services, housing and the administration of justice. The notice treats this list as non-exhaustive. Private entities providing similar services of a public nature should also carry out a FRIA. Private schools, universities, clinics and social service providers should therefore check their position carefully.
The FRIA and the provider’s obligations
The notice explains why the FRIA does not duplicate the provider’s work. Under Article 9, the provider must set up a risk management system. It covers only the risks that can reasonably be mitigated or eliminated through the design or development of the system, or through adequate technical information. The risks that arise from a specific use and context belong to the FRIA.
The provider’s instructions for use, which reflect its risk management system, are the starting point. Under Article 27(2), the deployer can also rely on:
- a previous FRIA;
- an impact assessment carried out by the provider, where the system and the risks are similar.
The provider is not required to share its assessment, but the notice calls sharing it a good practice. In negotiating contracts with providers, deployers should ask for it.
The FRIA and the DPIA
The FRIA complements the data protection impact assessment (DPIA). Where a DPIA has been carried out for the processing linked to the AI system, the deployer can cross-refer to it or incorporate the relevant parts. The template builds this in: a dedicated section on the right to the protection of personal data is designed to reuse the DPIA. The deployer then adds any further effects arising from the specific context of use.
When to carry out the FRIA, and when to update it
The FRIA must be carried out before the deployer first uses the system. It is needed for every AI system and must cover all its purposes of use. It must be updated when relevant factors change. The notice gives examples:
- the processes concerned;
- the persons affected;
- the version of the system or of its underlying model.
The last example is important. An update by the provider to the system or its underlying model may require the deployer to review its FRIA.
Deployers must comply with Article 27 from 2 December 2027. For the changes to the AI Act timeline, see The Digital Omnibus on AI: What Regulation (EU) 2026/1744 Actually Changes.
How to prepare
Before completing the template, the notice advises deployers to:
- designate, where possible, a multidisciplinary team that combines fundamental rights expertise, technical understanding of the system and operational knowledge of the context;
- assign roles clearly, including responsibility for the final assessment;
- obtain, review and keep the provider’s instructions for use and technical documentation;
- where appropriate, seek input from affected persons or groups, independent experts, authorities or civil society organisations.
The template, section by section
Section 1 – General information
- Identification: the deployer and the type of entity (public body, private entity providing public services, or credit/insurance deployer), and the system’s number in the EU database.
- Area of use: the Annex III area or areas in which the system is used.
- Status of the FRIA: whether it concerns the first use, relies on an earlier assessment, or updates a previous FRIA.
- Consultation: which stakeholders were consulted.
Section 2 – Context of use
- Purpose and process: the intended purpose stated by the provider, whether the use is in line with it, and how the system supports or automates the process.
- Duration and frequency of use.
Section 3 – Affected persons and groups
- Checklist: children, older people, women, LGBTIQ+ persons, persons with disabilities, persons in a vulnerable social or economic situation, persons with a minority racial or ethnic background, religious or belief groups, and third-country nationals including asylum seekers and refugees.
- The general public: may be ticked, but the template states that it does not replace the specific groups.
Section 4 – Risks to fundamental rights. This is the core of the template.
- Personal data: the first table covers the right to the protection of personal data and builds on the DPIA.
- Other rights: the deployer then describes each concrete risk of harm in its own block. It ticks the rights affected from a list based on the EU Charter of Fundamental Rights, which includes human dignity, non-discrimination, private life, education, workers’ rights, social security, good administration, effective remedy, the rights of the child and academic freedom. It also identifies the groups concerned.
- Rating: likelihood and severity are each rated low, medium or high. Mitigation measures are not required for low risks. The residual risk after mitigation must also be rated.
Section 5 – Human oversight
- Measures in place: which of the provider’s measures have been implemented, who carries out oversight, and what power they have to override, disregard or reverse outputs. Deviations from the instructions for use must be explained.
- People: the competence, training, authority and support of those assigned to oversight, including AI literacy measures under Article 4. On this topic, see also AESIA Guide No. 6: Human Oversight.
Section 6 – Measures if risks materialise
- Monitoring and response: monitoring, reporting and evaluation, including the suspension of the system and reports to the provider under Article 26(5).
- Complaints and governance: complaint handling, internal governance and, where available, redress mechanisms.
Section 7 – Overall conclusion
- Synthesis: a short summary of the risks identified. The automated tool may generate it.
Notification and publication
Once the template is completed or updated, the deployer must notify the market surveillance authority of its Member State and attach the completed template. Where an authority has exceptionally authorised a system under Article 46(1), for example for reasons of public security, the deployer may be exempt from notification. It must still carry out the FRIA.
Some deployers must also register a summary of the findings in the public EU database for high-risk AI systems: public authorities, EU institutions, bodies, offices and agencies, and those acting on their behalf. The template includes a dedicated section for this. It covers biometrics (except in law enforcement, migration, asylum and border control), education, employment, essential public services and benefits, and the administration of justice and democratic processes. Under Article 77, authorities protecting fundamental rights may also request access to the FRIA through the market surveillance authority.
How to take part in the consultation
Feedback on the template and the notice can be sent through an EU Survey until 15 January 2027. Deployers that will have to carry out a FRIA can also apply to test the template on one of their own systems, with the support of the AI Office, by 4 December 2026.
- Draft notice and template: download
- Commission news item: link
- Feedback: EU Survey
- Testing: expression of interest
What to do now
- Check whether you are in scope. Map your AI systems against Annex III, and check whether your organisation is a body governed by public law, a private entity providing public services, a lender or an insurer.
- Ask providers for documentation. Ask for the instructions for use and, where possible, the provider’s own impact assessment.
- Connect the FRIA to your DPIA. Identify what can be reused and which effects need a separate analysis.
- Plan for updates. Set up a process to review the FRIA when the system, its underlying model, the processes or the affected persons change.
- Organise oversight and complaints. The template asks who can stop or correct the system, and how affected people can complain.
- Take part. Testing or commenting on the template is a way to shape its final version.
A FRIA is not a form to fill in once. It assesses how a specific system, used in a specific context, can affect real people. The official template will make assessments more uniform across the EU, but the quality of the answers will still depend on the deployer.
Data Valley supports public bodies, companies and research institutions in carrying out fundamental rights impact assessments, coordinating them with DPIAs, and preparing for the obligations of deployers under the AI Act.
Sei pronto a trasformare i dati in valore per il tuo business?

