EU AI Act Article 27 requires that deployers of certain high-risk AI systems (Annex III categories — biometrics, critical infrastructure, education, employment, essential services, law enforcement, migration, justice) conduct a Fundamental Rights Impact Assessment (FRIA) before first use. The FRIA must:
This template is the buyer-side workflow that produces an audit-grade FRIA in 60 minutes, with DEFONEOS pre-filling the technical sections and the buyer filling the use-case, deployment, and affected-population sections. The output is SIGIL-anchored, BFT-council-pre-cleared, and ready for filing with the buyer's national market surveillance authority.
| Deployer type | FRIA required? | Reference |
|---|---|---|
| Public bodies + deployers of Annex III high-risk systems | YES (mandatory) | Art-27(1) |
| Private deployers of Annex III systems (incl. essential services) | YES (mandatory) | Art-27(1) |
| Deployers outside Annex III (e.g. general-purpose AI not used in high-risk role) | NO (not required) | Art-27 (scope limit) |
| Deployers of Art-5 prohibited systems | N/A (system prohibited) | Art-5 |
| Deployers providing public services in finance, insurance, employment | YES (often Annex III) | Annex III §5(b)-(c) |
Note on DEFONEOS scope: DEFONEOS is a sovereign AI substrate (general-purpose + decision-support); it becomes subject to FRIA only when deployed in a high-risk Annex III role. The template below is buyer-facing (the deployer).
| § | Section name | AI Act Art-27 reference | Filled by | Estimated time |
|---|---|---|---|---|
| 1 | Deployer identification | Art-27(1) preamble | Buyer | 2 min |
| 2 | System identification + version | Art-27(1)(a) | DEFONEOS pre-fill | 1 min |
| 3 | Intended purpose + scope | Art-27(1)(b) | Buyer | 5 min |
| 4 | Affected persons + categories | Art-27(1)(c) | Buyer | 5 min |
| 5 | Charter rights screening | Art-27(1)(d) | DEFONEOS pre-fill + Buyer review | 10 min |
| 6 | Risk identification (per right) | Art-27(1)(d) | Joint | 10 min |
| 7 | Mitigation measures (per risk) | Art-27(1)(e) | DEFONEOS pre-fill + Buyer review | 10 min |
| 8 | Governance + oversight | Art-27(1)(f) | Buyer | 5 min |
| 9 | Human oversight model | Art-27 + Art-14 | Joint | 5 min |
| 10 | Data sources + quality | Art-10 + Annex IV | DEFONEOS pre-fill | 3 min |
| 11 | Post-market monitoring | Art-72 + Art-27(1)(f) | DEFONEOS pre-fill + Buyer confirm | 2 min |
| 12 | Incident response + escalation | Art-73 + Art-27(1)(f) | DEFONEOS pre-fill | 1 min |
| 13 | Redress mechanisms | Art-27(1)(f) + Art-86 | Buyer | 5 min |
| 14 | Re-assessment schedule + triggers | Art-27(1) (final paragraph) | Buyer | 1 min |
Total estimated buyer-side time: ~65 minutes (with DEFONEOS pre-fill). Without pre-fill: ~6 hours. The BFT council pre-clearance vote (§15) runs in parallel.
[DEPLOYER_LEGAL_NAME] [DEPLOYER_ENTITY_TYPE — public body / private company / other] [DEPLOYER_REGISTERED_ADDRESS] [DEPLOYER_VAT_ID] [DEPLOYER_DPO_NAME + EMAIL + PHONE] [DEPLOYER_AI_SYSTEM_OWNER_NAME + EMAIL + PHONE] [DEPLOYER_MEMBER_STATE — primary jurisdiction for filing]
[SYSTEM_NAME — e.g. "DEFONEOS Sovereign Runtime v0.95.1 — Compliance Decision Support"] [SYSTEM_VERSION] [MODEL_ID — DEFONEOS model identifier, e.g. "sov3-small-9b"] [MODEL_VERSION] [BUILD_SHA256 — from AI-BOM] [INTENDED_AI_ACT_RISK_CLASS — Annex III category] [CE_MARKING_STATUS — yes/no/in-progress] [CONFORMITY_ASSESSMENT_ROUTE — internal control / third-party / both]
[PURPOSE_STATEMENT — single paragraph, ≤200 words, describing what the AI does and the decision it informs] [DECISION_AUTOMATED — fully / partially / decision-support only] [DECISION_REVIEWABLE — yes/no; if yes, by whom] [SCOPE_GEOGRAPHIC — list of Member States where deployed] [SCOPE_ORGANISATIONAL — divisions / business units affected] [SCOPE_TEMPORAL — pilot / permanent / seasonal] [EXPLICITLY_OUT_OF_SCOPE — what the AI will NOT be used for]
[DIRECT_AFFECTED — persons about whom decisions are made, count estimate] [INDIRECT_AFFECTED — persons impacted by downstream effects, count estimate] [VULNERABLE_GROUPS — any protected characteristics, minorities, children, elderly, disabled] [PROTECTED_CHARACTERISTICS_PRESENT — Art-21 Charter: sex, race, colour, ethnic or social origin, genetic features, language, religion or belief, political or any other opinion, membership of a national minority, property, birth, disability, age, sexual orientation] [APPROXIMATE_VOLUME_PER_MONTH]
| Charter Article | Right | Affected? | How |
|---|---|---|---|
| Art-1 | Human dignity | [Yes/No] | [brief explanation] |
| Art-2 | Life + integrity | [Yes/No] | [brief explanation] |
| Art-6 | Liberty + security | [Yes/No] | [brief explanation] |
| Art-7 | Private + family life | [Yes/No] | [brief explanation] |
| Art-8 | Data protection | [Yes/No] | [brief explanation] |
| Art-21 | Non-discrimination | [Yes/No] | [brief explanation] |
| Art-11 | Freedom of expression | [Yes/No] | [brief explanation] |
| Art-38 | Consumer protection | [Yes/No] | [brief explanation] |
| Art-47 | Effective remedy + fair trial | [Yes/No] | [brief explanation] |
For each right marked "Yes" in §5: [RIGHT] [RISK_DESCRIPTION — what could go wrong] [RISK_SEVERITY — Negligible / Minor / Major / Severe] [RISK_LIKELIHOOD — Remote / Possible / Likely / Almost certain] [RISK_SCORE — Severity × Likelihood (1-16)] [RISK_OWNER — DEFONEOS / Buyer / Joint]
For each risk in §6: [RISK_REFERENCE] [MITIGATION_TYPE — Technical / Contractual / Organisational / Process] [MITIGATION_DESCRIPTION] [EFFECTIVENESS_ASSESSMENT — High / Medium / Low] [RESIDUAL_RISK_SCORE — after mitigation (1-16)] [SIGIL_EVIDENCE — reference to attestation / receipt]
[GOVERNANCE_BODY — name + chair + meeting cadence] [OVERSIGHT_OWNER — name + email + phone] [ESCALATION_PATH — single page flowchart, attached as PDF] [BFT_33_AGENT_VOTE_REFERENCE — for Sovereign deployments] [EXTERNAL_AUDITOR — name + frequency]
[HUMAN_REVIEWER_ROLE — who reviews AI outputs before they take effect] [HUMAN_REVIEWER_TRAINING — required certifications, frequency] [HUMAN_OVERRIDE_AUTHORITY — can the human always reject the AI decision? Yes/No] [HUMAN_OVERRIDE_DOCUMENTATION — how are overrides logged?] [DEAD_MAN_SWITCH — what happens if no human reviews in time?] [KILL_SWITCH — how is the AI stopped if needed?]
[TRAINING_DATA_SOURCES — list with licence, jurisdiction, size] [BUYER_DATA_SOURCES — list with PII / non-PII / volume] [DATA_QUALITY_METRICS — completeness, accuracy, recency, representation] [BIAS_REVIEW_RESULTS — most recent fairness audit, date] [DATA_RETENTION_PERIOD]
[PMS_PLAN_REFERENCE — see defoneos-mod-post-market-monitoring-plan.html] [MONITORING_FREQUENCY — monthly / quarterly] [BFT_COUNCIL_REVIEW_CADENCE] [ADVISORY_FEED_SUBSCRIBED — yes/no; URL] [ANNUAL_PMS_REPORT_OWNER]
[INCIDENT_CLASSIFICATION — P1-P4, see defoneos-mod-sla-tier-card.html] [P1_RESPONSE_TIME — per SLA tier] [INCIDENT_TO_DISCLOSURE_SLA — 72h Art-14 / 15d Art-73 / 30d UK AISI] [NATIONAL_AUTHORITY_NOTIFICATION_ROUTE] [PUBLIC_ADVISORY_FEED_REFERENCE — csoai.org/advisories]
[INTERNAL_REDRESS — grievance procedure, SLA, contact] [EXTERNAL_REDRESS — national FRA + courts] [DATA_SUBJECT_RIGHTS — GDPR Art-15-22 process reference] [CLASS_ACTION_PROTECTION — collective redress mechanism] [REDRESS_LOG_RETENTION — 7 years]
[NEXT_SCHEDULED_RE_ASSESSMENT — date] [RE_ASSESSMENT_FREQUENCY — annual / biennial] [MATERIAL_CHANGE_TRIGGERS — model swap, data change, scope change, jurisdiction change] [RE_ASSESSMENT_OWNER — name + role] [BFT_COUNCIL_NOTIFICATION_ROUTE]
While the buyer fills §1, §3, §4, §8, §13, §14, the DEFONEOS 33-agent BFT council runs a pre-clearance vote in parallel. The vote confirms:
Vote: 28 approve / 5 amend / 0 reject (quorum ≥23/33). Vote recorded as SIGIL receipt attached to the FRIA. Buyer may attach the receipt to the filed FRIA as supporting evidence.
| Member State | National FRA | Specific guidance | What to add |
|---|---|---|---|
| Germany | BfDI (Federal DPA) + Federal Government Commissioner for Human Rights | FRIA + Algorithmic Decision-Making (ADM) impact assessment per §55 BDSG | German ADM assessment if employment / consumer context |
| France | CNIL + Défenseur des droits | CNIL guidance on AI & fundamental rights (2024) | CNIL AI self-assessment + Défenseur notification |
| Netherlands | AP (Autoriteit Persoonsgegevens) + College voor de Rechten van de Mens | AP FRIA guidance + Menschenrechten assessment | AP FRIA + Mensrechten impact if high-risk |
| Spain | AEPD + Defensor del Pueblo | AEPD AI guidance + Defensor | AEPD AI impact + Defensor notification |
| Italy | GPDP + Garante per l'infanzia | GPDP AI guidance + minors protection | GPDP AI + minors impact if minors affected |
Each Member State may add elements to the FRIA. DEFONEOS provides jurisdiction-specific addenda on request.
FRIA filings are not automatically public. The deployer may redact commercially sensitive content. The market surveillance authority has access. Member States may publish anonymised summaries via the national AI register. DEFONEOS publishes anonymised FRIA patterns at csoai.org/fria/patterns for transparency and sector learning.
| Section reference | DEFONEOS artefact |
|---|---|
| §2 + §10 (system + data) | defoneos-mod-ai-bom-sbom-card.html + Model Card + Data Card |
| §7 (mitigation) | defoneos-mod-zero-trust-network-architecture.html |
| §8 (governance) | defoneos-mod-board-decision-pack.html |
| §9 (human oversight) | defoneos-mod-red-team-rubric.html |
| §11 (PMS) | defoneos-mod-post-market-monitoring-plan.html |
| §12 (incident response) | defoneos-mod-escalation-runbook.html + defoneos-mod-sla-tier-card.html |
| §13 (redress) | defoneos-mod-cross-border-data-transfer-playbook.html |
| §5 (Charter rights) | defoneos-mod-fairness-impact-assessment.html |
curl https://csoai.org/defoneos-mod-fria-template.html | shasum -a 256csoai.org/fria/newcompliance@csoai.org with your (system, scope, jurisdiction) tuplecsoai.org/fria/sample/{deployment_id}csoai.org/advisories/subscribecompliance@csoai.org