Know what’s on your site before a plaintiff’s firm does
A free privacy audit shows which cookies, pixels and trackers are running — and which ones commonly appear in privacy claims.
On November 16, 2026, the European Commission put a draft of one of the AI Act’s most anticipated practical tools out for comment: the template that certain deployers of high-risk AI systems will use to carry out and notify a fundamental rights impact assessment, or FRIA, under Article 27. The draft comes with an explanatory notice, a public feedback window that runs to January 15, 2027, and a call for deployers willing to test the form on one of their own systems. The obligation applies from December 2, 2027, but the draft already shows what regulators will expect to see.
TL;DR
- The European Commission published a draft FRIA template and explanatory notice under Article 27(5) of the AI Act on November 16, 2026, with feedback open until January 15, 2027.
- Deployers who want to test the draft on a real system can express interest to the AI Office by December 4, 2026, and the FRIA obligation applies from December 2, 2027.
- The template follows the six elements of Article 27(1) and lets deployers cross-refer to an existing DPIA for the data protection part instead of starting from scratch.

What the Commission actually published
The Commission’s announcement describes two parallel exercises. The first is a stakeholder consultation on the draft template, which the AI Office prepared under Article 27(5) with detailed input from the EU Agency for Fundamental Rights. The second is an open call for deployers that will be subject to Article 27 to apply the draft to one of their own systems, with an introductory session, support along the way and a closing session with the AI Office. Participation is voluntary, and if interest is high the AI Office may select cases for a balanced mix of use cases, deployer sizes and Member States.
Two points of status are worth being precise about. First, this is a draft: the explanatory notice and template state that they have not been adopted or endorsed by the Commission and reflect the preliminary views of Commission services. Second, once finalised, the template will not be optional. Deployers will need to use it to prepare the FRIA and to notify the results to the market surveillance authority under Article 27(3). The Commission also plans to turn the final version into an automated tool on the AI Act Single Information Platform. Who receives notifications will depend on each Member State’s set-up, and national laws such as Greece’s AI Act framework law are now designating the competent authorities.
The December 2, 2027 date reflects the revised timetable for Annex III high-risk obligations. Organisations planning around the original August 2026 date have gained time, but not much.
Who has to do a FRIA
Not every deployer of a high-risk system is caught. The explanatory notice lists three groups:
- Bodies governed by public law deploying a system in any Annex III area except critical infrastructure. The notice reads this broadly, from State organs and local authorities to judicial bodies and EU institutions.
- Private entities providing public services, again in any Annex III area except critical infrastructure. Recital 96 gives education, healthcare, social services, housing and the administration of justice as examples, and the notice treats that list as illustrative.
- Credit and insurance deployers, meaning those using AI to evaluate creditworthiness or set credit scores (fraud detection excluded) and those using AI for risk assessment and pricing in life and health insurance, regardless of whether they are public or private.
Deployers of systems used in critical infrastructure, point 2 of Annex III, are expressly exempt. A bank running AI credit scoring and a council using AI to triage housing applications are both in scope; a grid operator is not.
Walking through the draft template, section by section
The template mirrors the structure of Article 27(1) and adds an opening and a closing block. Here is what each part asks for and what it implies for your evidence file.
- General information. The deployer’s identity and type, a named responsible person, the system’s EU database identification number, the Annex III areas of use, whether this is the first deployment, whether the deployer is relying on a previous FRIA or on an impact assessment carried out by the provider under Article 27(2), whether this is an update, and which stakeholders were consulted and how their input was used.
- Context of use (Article 27(1)(a) and (b)). The intended purpose as defined by the provider, whether your use is in line with it, the process the system supports or automates, how outputs are used in practice, and the period and frequency of use.
- Affected persons and groups (Article 27(1)(c)). A checklist that includes children, older people, women, LGBTIQ+ persons, persons with disabilities, people in vulnerable social or economic situations, minority racial or ethnic groups, religious or belief groups, and third-country nationals including asylum seekers and refugees. “The general public” can be ticked but expressly does not replace the specific groups.
- Fundamental rights impact and risks of harm (Article 27(1)(d)). Part 4.1 covers the right to data protection and is built around the DPIA. Part 4.2 asks for one block per concrete risk of harm, ticking the Charter rights affected, from dignity and non-discrimination to workers’ rights and the right to an effective remedy. Each risk gets a low, medium or high rating for likelihood and severity, mitigation measures (not required for low risks), and a residual risk rating.
- Human oversight (Article 27(1)(e)). Which oversight measures from the provider’s instructions for use have actually been implemented, who carries them out, what authority they have to override or reverse outputs, where you deviated from the instructions and why, and how overseers are trained, including AI literacy measures under Article 4.
- Measures if risks materialise (Article 27(1)(f)). Monitoring and response mechanisms, including the ability to suspend use and the duty to report to the provider under Article 26(5), plus complaint handling, internal governance and any redress routes.
- Overall conclusion. A short synthesis of the risks found for each right, followed by a dated and signed submission.
Public authorities and EU bodies that must register their use in the EU database under Article 49(3) also fill in a summary of findings that will be published there. That summary does not apply to systems registered under Article 49(4), which covers law enforcement, migration, asylum and border control.
Building the FRIA alongside an existing DPIA
The most useful practical signal in the draft is how it treats the data protection impact assessment. The notice describes the FRIA as complementing the DPIA: the DPIA looks at risks from personal data processing to data subjects, while the FRIA looks at the deployment’s impact on fundamental rights more widely. Where a DPIA exists for processing linked to the AI system, deployers may cross-refer to it, attach relevant sections or copy them in. Teams that already run mature DPIAs, whether under the GDPR or regimes such as Brazil’s LGPD, have a head start.
In practice, that suggests a sensible workflow:
- Start from the DPIA’s risk register. Part 4.1 asks you to restate the DPIA’s likelihood and severity conclusions on the template’s three-point scale. If your DPIA uses a five-point matrix, decide now how you will map it, and document the mapping once rather than arguing it per system.
- Then widen the lens. A DPIA will rarely capture risks such as an AI tool narrowing access to social benefits, chilling freedom of expression or undermining the presumption of innocence. Part 4.2 is where those go, and the notice points deployers to the Charter, its official explanations and FRA’s Charterpedia.
- Pull in the provider’s work. Providers must run a risk management system under Article 9, but Article 9(3) limits it to risks that can be addressed through design or technical information. Risks arising from your specific deployment context are yours to assess. The notice says providers are not obliged to share their own assessments, though doing so is good practice, so ask for it in procurement.
- Use the provider’s instructions for use as the oversight baseline. Section 5 is explicitly anchored to them, and any deviation must be explained.
This fits the direction of practical AI governance: one assessment process feeding several legal outputs instead of parallel documents that contradict each other.
Timing, updates and the scoring scale
The notice is clear that the FRIA must be completed before first deployment and must cover every system and all its purposes of use. It must be updated when relevant factors change, and the examples given are telling: the processes concerned, the persons affected, or the version of the system or its underlying model. For systems built on frequently updated foundation models, that could mean regular reassessment, so version tracking belongs in change management.
The draft also defines its risk scale: low means unlikely under foreseeable conditions, medium means plausible under some conditions, and high means highly probable or already observed in comparable deployments. That final clause matters. Incidents reported elsewhere, including through the kind of cross-border AI incident reporting some Member States are pushing for, could push a risk into the high band for everyone using similar tools.
One narrow exemption exists: where a market surveillance authority has exceptionally authorised a system under Article 46(1), the deployer may skip notification, but not the assessment itself.
What to do before January 15
- Inventory AI systems that fall within Annex III and flag those deployed by public bodies, public-service providers, lenders and life or health insurers.
- Run the draft template against one live or planned system as a dry run, even if you do not join the official test.
- Note where questions are unclear or duplicate DPIA content, and send those points through the Commission’s feedback survey.
- Add FRIA support, including sharing of the provider’s own impact assessment, to AI procurement checklists.
- Assign an accountable owner, because the template asks for a named responsible person and a signature.
FAQ
Is the template legally binding yet?
No. It is a draft that reflects preliminary views of Commission services. Once finalised, deployers covered by Article 27 will have to use it to prepare and notify their FRIA.
Can a private company be caught by the FRIA duty?
Yes. Private entities providing public services, such as in healthcare, education or housing, are covered, as are lenders using AI credit scoring and life or health insurers using AI for risk assessment and pricing.
Does a completed DPIA satisfy the FRIA?
Not on its own. The DPIA can be cross-referenced for the data protection section, but the FRIA also asks about other fundamental rights, affected groups, human oversight and response measures.
Do we need to redo the FRIA when the vendor ships a new model version?
Possibly. The notice lists a change in the system’s version or its underlying model as an example of a change that requires the FRIA to be updated.
A form that asks the right questions
This template is closer to a design review than paperwork. Answering it honestly forces a deployer to say who could be harmed, how likely that is, who can switch the system off and how people can complain. The Digital Omnibus debate has bought deployers time, and as discussions about remedies when algorithms harm show, the cost of skipping these questions tends to arrive later and larger. The organisations that use the next year to practise on real systems will find December 2027 a formality rather than a fire drill.
Put this into practice
Turn privacy guidance into working controls.
Captain Compliance helps teams discover tracking technologies, enforce consent, manage privacy requests and document the evidence behind every decision.