Login Get a Demo
01 Glossary · Governance and risk

What Is an AI Impact Assessment?

What an AI impact assessment documents, how to run one, and how it differs from the EU’s fundamental rights impact assessment and a GDPR DPIA.

The short answer

What is an AI impact assessment?

Last updated
Read time
4 min read
From
Cranium AI
Also called
AI system impact assessment, Algorithmic impact assessment
Share on LinkedIn

An AI impact assessment is a documented evaluation of how an AI system and its foreseeable uses could affect individuals, groups and society, for better and for worse.

Where a risk assessment scores what could go wrong, an impact assessment looks outward, at the people on the receiving end of the system’s outputs.

Key takeaways
  • ISO/IEC 42005:2025, published May 28, 2025, gives guidance on when and how to assess AI system impacts.
  • ISO/IEC 42001 lists AI system impact assessment as clause 6.1.4, separate from risk assessment in 6.1.2.
  • The EU’s fundamental rights impact assessment (Article 27) is mandatory, but only for certain deployers.
  • US federal agencies must complete an AI impact assessment before deploying high-impact AI (OMB M‑25‑21).
By the numbers Each figure links to its sourceAll sources 
02 What it contains

What does an AI impact assessment document?

Formats vary. These four elements draw on EU AI Act Article 27 and OMB M‑25‑21, the two texts that spell the contents out.

01

Purpose and context

The intended purpose and expected benefit, the process the system will be used in, and the period and frequency of use (Article 27(1)(a) and (b)).

02

Who is affected

The categories of people and groups likely to be affected in that specific context, not users in general (Article 27(1)(c)).

03

Harms and benefits

The specific risks of harm to those groups (Article 27(1)(d)). M‑25‑21 names privacy, civil rights and civil liberties. NIST MAP 5.1 asks for the likelihood and magnitude of each impact, beneficial or harmful.

04

Oversight and response

How human oversight is implemented and what happens if a risk materializes (Article 27(1)(e) and (f)), plus a schedule for reassessment, which M‑25‑21 adds.

03 Compare

AI impact assessment vs. FRIA vs. DPIA

For how these differ from a risk assessment, see AI risk assessment.

AI impact assessmentFundamental rights impact assessmentData protection impact assessment
SourceISO/IEC 42005:2025; ISO/IEC 42001 clause 6.1.4EU AI Act, Article 27GDPR, Article 35
StatusVoluntary guidanceLegal dutyLegal duty
Who does itAny organization developing, providing or using AIPublic bodies, private providers of public services, and deployers of credit scoring or life and health insurance pricing AIThe data controller, where processing is likely to result in a high risk
Looks atImpacts on individuals and societyFundamental rights of the people affected by one deploymentRights and freedoms affected by personal data processing
OverlapCan be integrated into AI risk managementMay cross-reference a DPIA that already meets an obligation (Article 27(4))Can be cross-referenced by a FRIA instead of repeated
04 How to run

How do you run an AI impact assessment?

  1. Decide whether one is needed

    Screen every proposed use at intake. Uses that make or shape decisions about people go on to a full assessment.

  2. Describe the system in context

    Record purpose, users, data, deployment setting and the decisions the output feeds.

  3. Identify affected groups

    Name them, and gather input from operators, users and affected communities, as the NIST Playbook suggests under GOVERN 4.2.

  4. Assess and document

    Record each impact with its likelihood and magnitude, the mitigation and the residual effect.

  5. Decide, then revisit

    Use the result in the go or no-go decision, and update it when a documented element changes, as Article 27(2) requires for a FRIA.

05 Keep going

Related reading.

06 Questions

Frequently asked questions.

Straight answers, kept current. Last updated .

Ask us anything →
Is an AI impact assessment legally required?

It depends on who you are. ISO/IEC 42005 is voluntary guidance. In the EU, Article 27 requires a fundamental rights impact assessment from certain deployers of high-risk AI before first use, under rules the Commission’s timeline applies from December 2, 2027. In the United States, OMB M‑25‑21 requires one of federal agencies for high-impact AI. It does not bind private companies.

Who has to carry out a fundamental rights impact assessment?

Deployers that are bodies governed by public law, private entities providing public services, and deployers of the Annex III systems in point 5(b) and (c): AI that evaluates creditworthiness or sets a credit score, and AI for risk assessment and pricing in life and health insurance. They notify the results to the market surveillance authority (Article 27(3)).

What is the difference between an AI risk assessment and an AI impact assessment?

A risk assessment scores what could go wrong with a system: likelihood, severity and treatment. An impact assessment documents who is affected and how, including benefits. ISO/IEC 42001 lists them as separate clauses, 6.1.2 and 6.1.4, and the results of one feed the other. Running both in one review avoids asking a team the same questions twice.

07 See it for yourself

From the definition
to the evidence.

See how Cranium AI discovers, observes, governs, secures and proves every AI model, agent and vendor in your enterprise.