Login Get a Demo
01 Glossary · Governance and risk

What Is an AI Audit?

Who audits AI, against what criteria, what comes out, and how an audit differs from AI assurance and a conformity assessment.

The short answer

What is an AI audit?

Last updated
Read time
4 min read
From
Cranium AI
Also called
Algorithmic audit
Share on LinkedIn

An AI audit is an independent, evidence-based review of an AI system, or of how an organization manages AI, against defined criteria such as a law, a standard or internal policy.

It ends in findings of conformity or nonconformity. An audit is one form of AI assurance, and the most formal one.

Key takeaways
  • An audit needs three things: an independent auditor, written criteria and evidence to test.
  • ISO/IEC 42006:2025 sets requirements for bodies that audit and certify AI management systems.
  • New York City requires a yearly bias audit, by an independent auditor, of automated hiring tools.
  • Internal auditors have their own reference: The IIA’s Artificial Intelligence Auditing Framework.
02 Types

What kinds of AI audit exist?

Who audits, and against what, decides how much weight the result carries.

AuditPerformed byCriteriaOutput
Internal auditThe organization’s internal audit functionInternal policy and control objectivesReport to the board or audit committee
Certification auditA certification body that meets ISO/IEC 42006:2025ISO/IEC 42001 requirementsCertificate for the AI management system
Bias auditAn independent auditorA legal test, such as selection rates and impact ratios under New York City Local Law 144Published summary of results
Compliance auditInternal or external reviewersInternal policies, regulations and legal requirementsFindings and corrective actions
03 How it works

How does an AI audit run?

The sequence is the same whether the subject is one system or a whole management system.

  1. Fix the criteria

    Name the standard, law or policy the audit tests against. Without criteria, a review is an opinion.

  2. Set the scope

    Decide whether the subject is one AI system, a group of use cases or the management system around them.

  3. Collect evidence

    Inventories, AI‑BOMs, test results, logs, policies and records of who decided what.

  4. Test and interview

    Sample the controls, repeat the checks and talk to the people who operate them.

  5. Report and follow up

    Record conformity, nonconformities and corrective actions, then verify the fixes.

04 Standards

What do standards and laws say about AI audits?

01

ISO/IEC 42001

The AI management system standard has an internal audit clause (9.2) next to management review (9.3), so auditing is part of running the system.

ISO 42001 certification
02

ISO/IEC 42006:2025

Published in July 2025, it supplements ISO/IEC 17021‑1 with AI-specific requirements for the bodies that audit and certify AI management systems.

ISO/IEC 42006:2025
03

New York City Local Law 144

Employers may rely on a bias audit of an automated employment decision tool for one year only, and must publish a summary of the results.

NYC DCWP guidance
04

Evidence, not narration

Cranium AI is not an auditor or a certification body. It keeps evidence an auditor can inspect: AI‑BOMs generated from code repositories and self-attestations for the EU AI Act, NIST AI RMF and ISO 42001.

AI Cards and evidence
05 Keep going

Related reading.

06 Questions

Frequently asked questions.

Straight answers, kept current. Last updated .

Ask us anything →
What is the difference between an AI audit and AI assurance?

AI assurance is the broader activity: the UK government describes it as measuring, evaluating and communicating the trustworthiness of AI systems. An audit is one assurance technique, set apart by an independent auditor, fixed criteria and a formal finding. Risk assessments, impact assessments and formal verification are others. Every audit is assurance; most assurance work is not an audit.

Who can perform an AI audit?

It depends on the purpose. Internal audit functions review AI governance and controls for the board. Certification bodies audit AI management systems against ISO/IEC 42001, and ISO/IEC 42006:2025 sets the requirements those bodies must meet. Some laws define the auditor: New York City says an auditor who works for the employer or the vendor is not independent.

Is an AI audit the same as a conformity assessment?

No. A conformity assessment demonstrates that a product or system meets the requirements that apply to it. The EU AI Act sets conformity assessment procedures for high-risk AI systems in Article 43, some based on the provider’s internal control and some involving a notified body. An AI audit is a broader term for independent review against any stated criteria.

What evidence does an AI auditor ask for?

Expect requests for an inventory of AI systems and their owners, a bill of materials for each system, risk and impact assessments, test and red-team results, runtime records, incident logs, vendor documentation and proof that policies were followed. Evidence that systems generate continuously carries more weight than documents assembled for the audit.

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.