BlogFAQ
What is an AI audit?
4 min read
Two things get called 'AI audits.' Don't confuse them. An internal AI audit is a self-assessment your team runs to find gaps. An external AI audit is performed by a third party against a defined standard, usually for certification, regulatory filing or enterprise procurement.
What an AI audit examines
Scope varies by framework, but the recurring areas are consistent.
- Governance — policies, named owners, escalation paths
- Inventory — every AI system, vendor and dependency
- Data — sources, consent, retention, residency
- Model — testing, evaluation, bias, robustness
- Vendor — third-party risk and contractual terms
- Oversight — where humans review or override
- Monitoring — production behaviour and incident response
- Evidence — exportable records mapped to framework requirements
Who performs the audit
Three options, in order of cost and authority:
- Internal team — quick, cheap, useful for finding gaps before anyone else does
- Specialist consultancy — expensive but pragmatic; good before a real external audit
- Accredited certification body — required for ISO 42001 certification or formal EU AI Act conformity assessment
Timeline and cost for best results
An internal audit on a small AI stack: 1–2 weeks of work for one person. A consultancy-led readiness audit: 4–8 weeks, $20k–$60k. A full external ISO 42001 audit: 6–12 weeks of audit activity, $25k–$50k in audit fees plus internal effort.