Comparison

ISO/IEC 42005 vs EU AI Act FRIA: Impact Assessment Comparison

ISO/IEC 42005:2025 is an international standard that gives organizations guidance for performing AI system impact assessments. The fundamental rights impact assessment (FRIA) under Article 27 of the EU AI Act is a legal obligation within Regulation (EU) 2024/1689. One assessment process can feed both, but the standard does not by itself establish compliance with Article 27. In either case, the assessment describes expected impacts. It does not constrain what a deployed system does at runtime.

Turning findings into runtime policy requirements

  1. 1

    1. Name the finding precisely

    State the affected system, tool, data class and the harm the finding describes.

  2. 2

    2. Write a testable requirement

    Express the mitigation as a condition that can be evaluated for a single interaction or tool call.

  3. 3

    3. Choose the enforcement action

    Decide whether the condition should allow, redact, block, or escalate for human approval.

  4. 4

    4. Assign ownership and versioning

    Record the policy owner, the policy version, and the assessment finding it implements.

  5. 5

    5. Define retained evidence

    Specify what is recorded, for how long, and who can export it.

  6. 6

    6. Feed changes back

    When a model, tool or data source changes, revisit the finding and the policy together.

Two instruments with different jobs

ISO/IEC 42005:2025, titled Information technology, Artificial intelligence, AI system impact assessment, is an international standard, not legislation. ISO describes it as guidance for organizations that carry out AI system impact assessments, and as a framework for identifying, evaluating and documenting the impacts of AI systems. It works at the level of the individual AI system.

ISO also positions it as a companion to ISO/IEC 42001, the AI management system standard. In ISO's own characterization, 42001 makes impact assessments a consistent and required part of AI governance, while 42005 makes them thorough, systematic and actionable. The standard is therefore best read as a method that supports a management system.

The FRIA is different in kind. It is an obligation set out in Article 27 of Regulation (EU) 2024/1689. The Commission states that the Act's provisions apply progressively, and that as of October 2026 the enforcement powers of the AI Office and national competent authorities apply from 2 August 2026, when certain provisions become enforceable. Whether and how Article 27 applies to a given deployment is a legal determination. It should be made by counsel for each system and recorded, not assumed from the presence of an ISO-aligned assessment.

The practical consequence is that the two cannot be treated as interchangeable. The standard tells you how to run a good assessment. The regulation defines what a specific legal duty requires. A mature program uses the first to produce work that can be adapted to the second.

Side-by-side comparison

The table covers the dimensions that can be stated without reference to the detailed text of either instrument. Article 27 specifics such as exact content elements, deployer scope and notification duties should be taken from the regulation itself when you build your FRIA fields.

  • Legal status: ISO/IEC 42005 is a published international standard that provides guidance. The FRIA is a legal obligation within an EU regulation, and enforcement powers of the AI Office and national competent authorities apply from 2 August 2026.
  • Unit of assessment: ISO/IEC 42005 addresses individual AI systems. A FRIA is likewise tied to a specific deployment, so both need a precise system description.
  • Governance context: ISO/IEC 42005 is designed to work with ISO/IEC 42001 at the organization level. The FRIA sits inside the legal framework of the AI Act.
  • Process framing: ISO describes the 42005 process as identifying, evaluating or analyzing, and documenting impacts. The FRIA is a defined legal assessment, so its content is fixed by the regulation rather than chosen by the organization.
  • Consequence of weak output: A weak ISO-aligned assessment is a management system gap. A weak FRIA is a potential compliance gap before a regulator.
  • Reuse: A shared assessment core can serve both. FRIA-specific fields should be added and approved separately, with legal review.

Building one assessment process that supports both

The most defensible design is a single intake with a shared core and a clearly marked FRIA layer. The core holds what any competent impact assessment needs: a system description, the affected parties, the impact analysis, and the mitigations. This content maps naturally to what ISO/IEC 42005 describes, and it is the material a FRIA draws on.

The FRIA layer holds everything the regulation requires beyond that core. Keep it as separate fields with a named legal owner, so that no one mistakes completion of the ISO-style core for completion of the legal assessment. Do not claim that following ISO/IEC 42005 satisfies Article 27 or gives any presumption of conformity. The Commission says harmonised standards offer legal certainty under the Act, but nothing indicates that ISO/IEC 42005 is one.

Assign owners across risk, legal, privacy and engineering, and retain approval and review records. Define reassessment triggers explicitly: a system change, a new use case, an incident, or a change in models, tools or data sources. Take any mandated update timing from the standard and the regulation, and treat your own triggers as additions rather than substitutes.

The main tradeoff is effort against clarity. A single process avoids duplicated analysis and conflicting records. It also requires discipline about which fields are optional guidance and which are legal requirements.

Evidence that supports documentation, and its limits

Auditors and regulators generally want to see two things: what the organization decided before deployment, and what the system actually did afterward. The assessment answers the first. Runtime records answer the second. Reasonable evidence types include the approved assessment with its version history, the policy set derived from it, decision records showing which policy applied to which interaction or tool call, escalation and human approval records, and change records showing reassessment.

The traceable link matters more than any single artifact. For each finding, you should be able to show the policy that enforces it, the version in force at the time, and the decisions that policy produced. That chain lets one body of work support both ISO-style documentation and FRIA-style documentation.

Runtime evidence has limits. It is operational proof of behavior and can support documentation obligations, but it does not replace the assessment itself. A complete set of enforcement records cannot stand in for an analysis of affected parties and rights. Likewise, a thorough assessment without enforcement leaves a gap between what was predicted and what can be demonstrated.

Where Trussed fits

Trussed does not perform impact assessments and is not a GRC platform. It is runtime control and evidence infrastructure for regulated enterprises. Assessment findings that you have turned into policy requirements can be enforced in the execution path, across models, agents, tools, MCP servers, AI sites and coding assistants. Policy is enforced at the level of individual interactions and tool calls, with actions that include allow, redact, block, rate limit, and escalate for human approval, at runtime rather than reviewed after the fact.

That enforcement generates continuous, audit-ready evidence, mapped to frameworks including the EU AI Act and NIST. Retention and data residency are configurable, and deployment is self-managed in the customer cloud. The evidence supports the documentation your assessment process produces. It does not substitute for the assessment, and it does not determine whether a FRIA is required.

From assessment to enforced control

Assess

Identify, evaluate and document impacts for each AI system.

Translate

Convert each finding into a testable policy requirement.

Enforce

Apply allow, redact, block or escalate decisions in the execution path.

Retain

Keep decision records that can be shown to an auditor or regulator.

Connect assessment findings to enforcement and evidence

See how policy derived from your impact assessments is enforced on individual interactions and tool calls, and what audit-ready evidence that produces.

Explore Agent and Tool-Call Enforcement