FRIA under the AI Act: not (only) a public-sector affair, and the deadline has already moved once

On 16 November 2026, the European Commission opened a public consultation on the draft template — and accompanying explanatory notice — for the Fundamental Rights Impact Assessment (FRIA), the tool required under Article 27 of the AI Act for deployers of certain high-risk AI systems. This is not a public-administration-only obligation, and the 2027 date in the headlines is not the original one: between a reform that just rewrote the rules on reusing an existing DPIA and a postponement that pushed the compliance deadline forward, the preparation window is different — and narrower — than it looks at first glance.
At a glance
The dates that matter
- Who is obligated: not only public bodies, but also private entities providing public services, and deployers — public or private — of systems used for credit scoring (Annex III, point 5(b)) and for risk assessment/pricing in life and health insurance (Annex III, point 5(c)).
- FRIA is not a relabelled DPIA: it looks at fundamental rights, not only data protection. The Digital Omnibus has made reuse of an existing DPIA explicit, through cross-references or incorporated parts — not just "synergy".
- Do now: check whether you fall within scope, map which existing DPIAs are reusable, consider applying for the test, and don't wait until 2027 to draft a first internal FRIA.
What happened on 16/11 and why it matters
The Commission put out for consultation the draft FRIA template together with the explanatory notice that accompanies its use. The document was developed by the AI Office under Article 27(5) of the AI Act, with input from the EU Agency for Fundamental Rights (FRA). Once adopted, the template won't remain a simple form to fill in: it will be set up as an automated tool on the AI Act Single Information Platform, the single platform designed to manage the Act's information obligations. It is still a draft, open to change based on consultation feedback — but the fact that it is already moving into operational testing signals the Commission treats it as mature, not a preliminary exercise.
Who is obligated: not just the public sector
Article 27 of the AI Act identifies three categories of deployers required to carry out a FRIA, and only the first matches the common intuition. Obligated are: public bodies and entities governed by public law; private entities providing public services; and deployers — regardless of whether they are public or private — of high-risk systems falling under Annex III, point 5, letters (b) and (c): respectively, creditworthiness assessment and credit scoring (with an exception for anti-fraud systems, which are excluded) and risk assessment or pricing in relation to life and health insurance, with no equivalent exception. This last category is the point many private operators — banks, insurers, fintechs — risk missing: the obligation does not depend on being a public body, but on the type of system in use. The obligation always falls on the deployer — the entity using the system — never on the provider that developed it.
What the template actually asks for
In substance, Article 27(1) requires the deployer to describe, for each intended use: the deployer's processes in which the system will be used, matching its intended purpose; the period of time and frequency in which the system is intended to be used; the categories of natural persons and groups likely to be affected by the specific use; the specific risks of harm likely to impact those categories of persons; the human oversight measures to be implemented, following the provider's instructions for use; and the measures to be taken in case those risks materialise, including internal governance arrangements, complaint-handling included. Once completed, the assessment must be notified to the relevant market surveillance authority, together with the filled-in template. The detailed section-by-section breakdown reported by some industry trackers cannot yet be confirmed against the official template text, which — as noted — remains under consultation.
FRIA and DPIA: explicit reuse, not just synergy
Here the draft lands at a moment of changing rules, not stability. Until a few months ago, Article 27 provided that the FRIA should "complement" a DPIA already carried out for the same processing: the two assessments remained conceptually distinct, even when covering the same system in part. The Digital Omnibus — Regulation (EU) 2026/1744, in force since 27 July 2026 — rewrote paragraphs 4 and 5 of Article 27: deployers can now explicitly include cross-references to a DPIA, or incorporate relevant parts of it directly into the FRIA, to avoid duplicating the same work twice. This is a concrete simplification, not just a statement of principle. One distinction survives the reuse, though: a DPIA assesses risks to personal data protection, while a FRIA assesses impact on fundamental rights as a whole — a broader scope that also covers issues such as non-discrimination or access to essential services, which are not simply an extension of a DPIA.
The timeline that matters right now
Three dates, three different actions. Anyone wanting to apply for the voluntary template test — a selection designed to cover a balanced mix of use cases, deployer types and sizes, and Member States — must express interest to the AI Office by 4 December 2026. The public consultation on the draft remains open for comments, via EU survey, until 15 January 2027. The compliance obligation for deployers kicks in from 2 December 2027. This last date deserves a clarification that coverage of the consultation tends to skip: according to a convergence of professional sources on the Digital Omnibus, this is not the AI Act's original deadline, which for Annex III systems was set at 2 August 2026. The postponement of more than a year arrived together with the rewrite of the DPIA/FRIA paragraphs described above. The calendar, in other words, has already moved once in 2026 — and there is no guarantee it won't move again.
Takeaway: the operational checklist
First: check whether you fall within the scope of Article 27 — not only as a public body, but also as a private entity providing a public service, or as a deployer of credit-scoring or life/health insurance risk-assessment systems. Second: map which existing DPIAs already cover, even partially, the same systems, and where they stop short of the FRIA's broader scope. Third: consider applying for the voluntary template test by 4 December 2026, a way to arrive prepared before the template is finalised. Fourth, and most important: don't treat 2 December 2027 as a distant horizon. It is already a one-time extension; it's worth drafting a first internal FRIA now, using the current draft as a reference, rather than waiting for the final version to start.
Do you fall within the scope of Article 27?
We classify your high-risk AI systems, check whether the FRIA obligation applies and map the DPIAs you can already reuse, so your first draft doesn't start from scratch. Let's talk.
Let's talk about your FRIA →