An AI auditor is a professional who independently checks an AI system, or the organization running it, against set criteria such as a standard, a law or a contract. In the US the title covers three different jobs. Each audits something different, works to different criteria and starts from a different background.
The first job audits the management system around AI against ISO/IEC 42001:2023, the international standard for an AI management system (AIMS). The second tests models and their outcomes, for example the bias audits New York City requires for hiring tools. The third is any auditor who uses AI tools in an ordinary financial or IT audit. This guide explains each one and the routes in. It also shows where the ISO/IEC 42001 Lead Auditor credential from the Global AI Certification Council (GAICC) fits.
Three jobs called AI auditor
The three AI auditor jobs differ in what they examine, so pick the one that matches the work you want before you pick a credential. The table sets them side by side.
| Management system auditor (ISO/IEC 42001) | Algorithmic or model auditor | Auditor who uses AI tools | |
| What is audited | The organization’s policies, risk and impact assessments, controls and records for AI | A specific model or automated decision tool and its outputs | Financial statements, IT controls or operations, with AI used as an audit aid |
| Typical criteria | ISO/IEC 42001 clauses 4 to 10 and Annex A; ISO 19011:2026 for method | A law’s test (such as NYC impact ratios), a validation plan, fairness metrics | Auditing standards for the engagement; the firm’s own AI use policy |
| Output | Graded findings and an audit report; for certification, a recommendation to the body | A test report or bias audit summary | The normal audit opinion or report |
| Who hires | Certification bodies, in-house audit functions, consultancies, buyers auditing suppliers | Specialist audit firms, model risk teams, employers using hiring tools | Audit and accounting firms, internal audit functions |
| Typical background | ISO, GRC, IT or internal audit | Statistics, data science, model validation | Accounting, financial audit, IT audit |
Only the first job has a certification scheme behind it. Organizations get certified to ISO/IEC 42001 by accredited certification bodies, and people hold auditor credentials such as GAICC’s. Model audits and AI-assisted audits have no single standard yet. The rest of this guide covers the first two jobs, because the third is an existing audit career with new tools.
The difference between the first two jobs is easy to miss. A management system auditor asks whether the organization controls its AI, using sampled records as proof. A model auditor asks whether one system behaves within limits, using tests on that system. A Lead Auditor working to ISO/IEC 42001 does the first job, as set out in GAICC’s guide to the Lead Auditor exam and levels.
Audit criteria in force in the US in October 2026
AI auditors in the US work against a short list of current documents, and several older ones that many guides still cite are gone. Check the status line before you quote any of them to a client or an interviewer.
| Document | What it asks for | Status on 20 October 2026 |
| ISO/IEC 42001:2023 | Requirements for an AIMS; certifiable | Current; edition 1, December 2023 |
| ISO 19011:2026 | Guidelines for auditing management systems | Current; fourth edition, 27 May 2026 |
| NIST AI Risk Management Framework | Voluntary functions to govern, map, measure and manage AI risk | Current; version 1.0, January 2023 |
| NYC Local Law 144 | A bias audit of an automated employment decision tool within one year of use, plus notices | Enforced since 5 July 2023 |
| Colorado SB 26-189 | Notices, records and human review for automated decisions | Duties apply from 1 January 2027; no audit or framework duty |
| Federal Reserve SR 26-2 | Revised model risk management guidance for banks | Issued 17 April 2026; supersedes SR 11-7 and excludes generative and agentic AI |
| Executive Order 14110 (2023) | Federal AI safety order | Revoked by Executive Order 14148 on 20 January 2025 |
| EEOC 2023 technical assistance on AI in hiring | Adverse impact guidance for software and AI | No longer on the agency’s website; the page returned “Page not found” on 20 October 2026 |
Federal anti-discrimination law itself still applies to AI used in hiring. Only the agency’s explanatory document is gone. Bank model auditors should note the SR 26-2 scope line. The guidance attached to the letter calls generative and agentic AI “novel and rapidly evolving” and leaves both outside its scope. Those systems fall back on a bank’s wider risk and governance practices, which is where an AIMS can help. The bank-side detail is in model risk rules for US banks.
State rules change fastest. Colorado’s 2024 law named ISO/IEC 42001 and the NIST framework, and its 2026 replacement names neither. GAICC tracks those changes in how state AI laws treat ISO 42001. For EU work, high-risk duties under Regulation (EU) 2024/1689, as amended in 2026, now apply from 2 December 2027 for Annex III systems and 2 August 2028 for Annex I. Most Annex III systems follow internal control, so no outside body assesses them.
Routes into each kind of AI audit
Most AI auditors arrive from a neighboring audit or technical job, and the shortest route depends on what you already do. The routes below are GAICC editorial analysis.
| Your background | Shortest route | What you add first |
| ISO/IEC 27001 or ISO 9001 auditor | ISO/IEC 42001 management system audits | AI risk and impact assessment, Annex A controls for AI |
| Internal or IT auditor | AI audits inside your audit plan, then AIMS internal audits | AI life cycle, data provenance, model change control |
| GRC or compliance lead | Internal audit of the AIMS your employer runs | Audit method under ISO 19011:2026, evidence sampling |
| Data scientist or model validator | Model audits, bias audits, technical expert on AIMS audits | Audit independence, written findings, management system basics |
| Lawyer or privacy professional | Regulatory audits and impact assessment review | Testing evidence, ISO/IEC 42001 clause structure |
| No audit or technical background | A first role in GRC or internal audit | Audit basics before any AI credential |
The no-background route is the slow one. Employers hiring for AI audit work ask first for audit experience, and a credential without it rarely opens the door. Start where audits already happen and add AI scope to that work.
Skills and credentials by route
Each route has its own skill set and credential signal, and only the management system route has a family of 42001-specific credentials. Skills come first. Every route needs audit method, evidence handling and written findings. The AI part changes by route.
| Route | Core skills | Credential signals |
| Management system auditor | ISO 19011:2026 principles, audit planning, sampling, grading findings, ISO/IEC 42001 clauses and Annex A | GAICC Internal Auditor (Level 1), Lead Auditor (Level 2) and Senior Lead Auditor (Level 3); a certification body’s own approval for third-party work |
| IT or internal auditor of AI | Control testing, IT general controls, AI governance frameworks, data analytics | An IT audit certification; an IT audit association’s AI audit credential that requires an existing IT audit, internal audit or accounting designation |
| Model or bias auditor | Statistics, fairness metrics, model validation, documentation review | Specialist firms’ algorithm audit programmes; no single dominant credential |
GAICC’s management system credentials share one format. The Internal Auditor and Lead Auditor each need 32 contact hours of training and an exam of 60 questions in 90 minutes, with a 70% pass mark. The Senior Lead Auditor needs 40 contact hours and an 80-question exam. The GAICC Handbook describes the Lead Auditor as leading second-party audits and first certification audits of an AIMS, which is the management system job in the table above.
Certification body work adds one more gate. ISO/IEC 42006:2025 sets floors for every AIMS auditor a body uses. Each needs four years of full-time work in IT or data protection with at least one year related to AI systems, and 24 hours of AIMS audit training. The body checks these itself, as described in how a certification body approves AIMS auditors.
How an AIMS audit differs from a model audit
An AIMS audit tests whether an organization governs its AI, while a model audit tests whether one AI system performs within limits. Researchers make the same split. A 2023 paper by Mökander and colleagues on auditing large language models proposes governance audits, model audits and application audits as layers that “complement and inform each other.”
| Question | AIMS audit (ISO/IEC 42001) | Model audit |
| Unit of audit | The organization’s AIMS within its stated scope | One model or automated decision tool |
| Main evidence | Policies, risk and impact assessments, Statement of Applicability, records sampled over time | Test datasets, metrics, model documentation, outputs |
| How evidence is gathered | Interviews, observation and record sampling under ISO 19011:2026 | Statistical tests, benchmark runs, reproduction of results |
| What a gap looks like | A nonconformity against a clause or control | A metric outside a threshold, or an undocumented limitation |
| Output | Audit report with graded findings and, for certification, a recommendation | Test report or bias audit summary |
| Who decides what happens next | The certification body’s decision maker, or management for internal audits | The client, or a regulator where a law applies |
An AIMS audit hands over a fixed set of outputs. The auditor plans against the clauses and Annex A and samples evidence. Findings are graded as major or minor nonconformities or opportunities for improvement. The report records the evidence, findings and conclusion. For certification, the team recommends a decision, and someone who did not audit makes it. The grading rules and report contents are covered in how findings and the audit report are written.
The two audits feed each other. An AIMS audit checks that model testing happens and that its results reach the risk assessment. A model audit supplies the evidence that such testing exists. An employer with high-impact AI may need both.
Independence inside and outside the organization
AI audit work splits into internal and external tracks, and independence is managed differently on each. Internal AI auditors sit in the risk, compliance or internal audit function. They have ongoing access to systems and people, which gives depth, but they report up through the structure they audit. ISO/IEC 42001 handles that through clause 9.2, which asks for auditors chosen to keep each audit objective and impartial.
External auditors bring independence by design, and certification bodies have the strictest rules. ISO/IEC 42006 bars a body from consulting on AI management systems for its certification clients or running their internal audits. NYC Local Law 144 also asks for an independent auditor for each bias audit.
The hardest part of AI auditing is usually institutional. Bias testing tools are widely available. Getting a development team to produce records it never budgeted for is harder. So is keeping an escalation path that senior managers use, and protecting a finding when the system earns revenue. An auditor who can read the technical evidence and also hold a finding under pressure is the one employers describe.
Research supports that caution. A 2024 review of AI audit studies by Birhane and colleagues found that “only a subset of AI audit studies translate to desired accountability outcomes.” An earlier internal audit framework, SMACTR, set out five stages for internal teams: scoping, mapping, artifact collection, testing and reflection.
AI audit jobs in the US
US AI audit jobs rarely carry the title “AI auditor”. They appear as IT audit, internal audit, governance and certification body roles with AI or ISO/IEC 42001 written into the duties. No government occupation code covers it; federal labor statistics carry no AI auditor category.
Four ads open on 20 October 2026 show the pattern. An IT audit manager post at a consulting firm valued the 42001 Lead Auditor credential. A director of IT internal audit at a software maker would lead audit work behind the employer’s own 42001 and 27001 certificates. A telecom carrier sought a principal to run cybersecurity audit, risk and governance for AI and emerging technology, with its AI risk framework aligned to ISO 42001 and the NIST framework. And an advisory firm with an accredited certification arm wanted a senior associate to lead ISO/IEC 42001 certification and internal audits alongside security and privacy standards.
Model audit roles show up in model risk management, data science and specialist audit firms. Bias audits of New York City hiring tools are a steady source, and AI in hiring and the bias audit rules sets out what the city expects of the employer and the independent auditor.
Most of those ads asked for audit experience first and treated any AI credential as a plus. Titles to search on a job board include IT auditor, internal audit manager, technology risk, AI assurance, model risk and ISO auditor, each paired with the standard’s number. A search for the words “AI auditor” alone also returns thousands of AI engineering and architecture ads, so add “ISO 42001”, “audit” or “model validation” to narrow it.
Employer-posted pay for these ads, dated and matched to how each ad uses the standard, sits in dated US pay ranges for Lead Auditor roles. The same guide covers resume lines and interview questions for audit roles.
Where the ISO 42001 Lead Auditor fits, and next steps
The ISO/IEC 42001 Lead Auditor fits the management system job, and it is the credential that signals readiness to lead an AIMS audit. It does not certify model testing skills, and it does not by itself let you audit for a certification body. A practical path runs in four stages.
- Build AI literacy. Learn how models are trained, tested and changed, enough to question an engineer about data and monitoring.
- Learn the standard. GAICC’s Foundation level gives a grounding in ISO/IEC 42001 in 16 training hours, with no prerequisites.
- Earn the audit credential. Take the Internal Auditor if you will audit your own employer, or the Lead Auditor if you will lead audits of others. The GAICC ISO/IEC 42001 Lead Auditor programme covers the method and prepares you for the exam.
- Log real audits. Join internal audits, supplier reviews and certification body audits as a trainee, and record each one.
After that, two directions open. Third-party work means approval by a certification body, and later a team leader role. The Senior Lead Auditor credential suits auditors who lead large engagements across several sites and standards. Some auditors also add model audit skills and act as the bridge between the two jobs.
Each track has its own ladder. The steps below are GAICC editorial analysis, drawn from the titles in current US ads and the role rules in ISO/IEC 17021-1 and ISO/IEC 42006.
| Track | Early role | Lead role | Senior role |
| Certification body | Auditor in training, then authorized auditor | Audit team leader | Report reviewer or certification decision maker |
| Internal audit | IT or technology auditor | Audit manager for AI and technology risk | Director of IT audit, chief audit executive |
| Advisory | Senior associate on AIMS audits and readiness work | Manager of AI assurance engagements | Partner or principal for AI assurance |
| Model audit | Model validator or data scientist on audits | Lead validator or bias audit lead | Head of model risk |
Moves between tracks are possible. A certification body auditor can move in-house to own an employer’s AIMS audit programme, and an internal auditor can move to a certification body once the supervised audits are logged. Keep one audit log across all of them, because every track asks for the same proof of what you audited and in what role.
Frequently Asked Questions
Is AI auditing a licensed profession in the US?
No. There is no federal license for AI auditors, and none of the state AI laws GAICC reviewed licenses the title. Some laws require audits, such as New York City’s bias audits for hiring tools, and set rules for who may perform them. Certification bodies apply their own competence rules to the auditors they approve.
Do I need a computer science degree to become an AI auditor?
No. Management system auditors come from audit, risk, GRC and compliance work, and ISO/IEC 42001 audits rely on records and interviews more than code. Model auditors need more statistics and data science. For certification body work, ISO/IEC 42006 asks for education at university level or equivalent training, which need not be in computer science.
Is an AI audit the same as an AI ethics review?
No. An ethics review usually advises a team on whether a system fits stated values. An audit tests evidence against fixed criteria and records findings. An ISO/IEC 42001 audit grades nonconformities and expects corrective action, while an ethics committee’s view carries no such weight. Many organizations run both, with the ethics review feeding the impact assessment.
Can one person audit both the AIMS and the models?
Sometimes, but rarely alone on the same engagement. Under ISO/IEC 42006, an ISO/IEC 42001 certification audit team must collectively know every Annex A control, and a certification body can add technical experts for model topics. Technical experts support the auditors and do not act as auditors. In internal audit, one person can cover both if the plan and the skills allow.
How do ISO/IEC 42001 and the NIST AI RMF relate for an auditor?
ISO/IEC 42001 is a certifiable requirements standard, so an auditor can find a nonconformity against it. The NIST AI RMF is voluntary guidance with no certification. US employers often run the NIST framework as their operating model and use ISO/IEC 42001 for a third-party certificate. The mapping is covered in GAICC’s walk-through of the NIST AI RMF.
Which US sectors use AI auditors most?
Financial services, technology companies and employers using automated hiring tools are where the ads GAICC read and the rules above point. Banks have long had model risk and validation teams. Technology firms pursue ISO/IEC 42001 certification for customers. New York City employers need bias audits.
