FRIA template — EU AI Act Article 27

The EU AI Act (Regulation (EU) 2024/1689) Article 27 requires certain deployers of high-risk AI systems to perform a Fundamental Rights Impact Assessment (FRIA) before deployment. This page provides Attestix's structured FRIA template — 12 sections aligned to Article 27(2), with deterministic completeness checks, an ImpactAssessmentCredential Verifiable Credential wrapper, and an optional Base L2 Sepolia anchor for tamper-evident retention.

Why this matters

Who must perform a FRIA (per Art 27(1)):

1. Deployers that are bodies governed by public law.
2. Private operators providing public services.
3. Deployers of high-risk AI systems in Annex III points 5(b) and (c) — credit scoring and life/health insurance pricing.

The deployer is solely liable for the substance of the FRIA. The provider remains solely liable under Articles 16-22 for the AI system itself. The FRIA notification is submitted to the deployer's market surveillance authority per Art 27(3); it is complementary to (not a replacement for) the GDPR Art 35 DPIA, per Art 27(4).

Attestix does not write your FRIA. Attestix provides a structured, signable, anchorable template that produces a cryptographically-verifiable artefact your compliance team completes. When the EU AI Office publishes its official FRIA template, this template will be updated to align field-for-field with that publication.

The 12 sections (with worked VibeTensor example)

The table below shows each FRIA section, its mapped Art 27(2) sub-clause, the structured field shape, the deterministic check Attestix runs, and a worked example using VibeTensor as the deployer.

What the deterministic checks produce

Each FRIA section emits a security_check_id discriminator tag of the form eu_ai_act.art.27.fria_section_<N>_*. Failing a check produces a signed negative VerifiableCheckResult — signed evidence that the check did not pass, not silence. The deployer can still sign + submit a FRIA with checks failing (the checks are advisory, not blocking) — the failures appear prominently in the analyst-facing output so the deployer's compliance officer knows what is unresolved.

Auto-flag rules

Two deterministic auto-flags fire when an emerging risk profile demands compliance-officer attention:

1. High-likelihood × high-magnitude risk identified. Any §4 risk where likelihood ≥ likely AND magnitude ≥ major emits eu_ai_act.art.27.fria_section_4_high_likelihood_high_magnitude_flagged.
2. Unmitigated high-risk. Any §4 risk satisfying flag 1 that has no §6 mitigation linked emits eu_ai_act.art.27.fria_section_6_unmitigated_high_risk_flagged.

Neither flag blocks sign-off. Both surface visibly in the public bundle export so a reviewing auditor or regulator can see the deployer's risk-acceptance posture.

Important caveat — generate-fria CLI ships in v0.5.0. As of Attestix v0.4.0 the template structure is documented (this page); the structured CLI + MCP tool (attestix generate-fria --profile <id> --output fria.{pdf,json} and mcp__attestix__generate_fria) ship in v0.5.0. Until then you can use this page's structure as a checklist and apply your own signing.

What we don't do

• We do not author your FRIA content. The substance — risk identification, mitigation selection, governance arrangements — is the deployer's compliance team + general counsel. We provide the structure.
• We do not judge whether your mitigations are adequate. The §4 / §6 completeness check confirms the link; we never assert the substance of a mitigation is sufficient.
• We are not a substitute for the official EU AI Office template. When the EU AI Office publishes its template, this template will be updated to align field-for-field.
• We do not file the Art 27(3) notification on your behalf. Submission to your market surveillance authority is the deployer's act. We record the date + reference (§9).
• We are not your DPIA tool. The DPIA under GDPR Art 35 is its own assessment; the FRIA is complementary. We track the linkage (§8) but do not generate the DPIA itself (OneTrust, BigID, and similar privacy platforms do that).
• We do not auto-promote assertion_recorded to passed=true. Sections containing deployer-asserted narratives (§10, §11) are signed but never marked verified.

How to verify our coverage yourself

Python / CLI

# Generate a FRIA template for a compliance profile (v0.5.0+)
attestix generate-fria --profile <profile-id> --output fria.json

# Render as PDF for sign-off (v0.5.0+)
attestix generate-fria --profile <profile-id> --output fria.pdf

# Verify a signed FRIA credential (today; once you have a hand-signed VC)
attestix verify-credential fria-credential.json

# List all FRIA-related audit events for a profile (v0.5.0+)
attestix audit list --security-check eu_ai_act.art.27.fria_section_12_signed

JavaScript / browser

npm install attestix

import { verifyCredential } from "attestix";

const result = await verifyCredential(impactAssessmentCredentialJson);
// result.valid === true if the Ed25519 signature over the JCS-canonical body
// matches the issuer DID's public key.

On-chain anchor (Base L2 Sepolia testnet)

# Anchor the FRIA credential hash to Base L2 Sepolia for tamper-evident retention
attestix anchor credential fria-credential.json --network base-sepolia

Mainnet schema registration is planned; testnet is the default target today.

Comparable disclosure

How other tools handle FRIA / impact assessments.

See also

• OWASP Top 10 for Agentic Applications mapping
• ISO/IEC 42001:2023 mapping — Annex A.5 impact-assessment controls cross-walk to FRIA
• NIST AI RMF 1.0 mapping — MAP-5.1 likelihood/magnitude cross-walks to FRIA §4
• SOC 2 Trust Services Criteria mapping — CC3.2 risk-assessment (2022 AI/ML POF) cross-walks to FRIA §4/§6
• EU AI Act compliance guide
• The internal template spec at attestix-cloud-plan/26-FRIA-TEMPLATE.md.

Attestix is evidence tooling for deployers subject to EU AI Act Article 27. Attestix does not write the FRIA, does not judge whether the deployer's risk-mitigation measures are adequate, and does not substitute for the published EU AI Office template (when available). The deployer remains solely liable for the substance of the FRIA; the provider remains solely liable under Articles 16-22 for the AI system itself.