Auditability

Auditability is the ability for someone to examine an AI system against requirements using evidence. An audit asks whether the system did what it was supposed to do, whether controls worked, and whether exceptions were handled responsibly.

Auditability requires more than transparency. A system can expose many details while still making it hard to judge compliance. Auditable evidence is organized around requirements, samples, tests, logs, owners, and corrective actions.

Prepare for an Independent Audit

An automated benefits system is under audit. For each requirement, pick the evidence from the bank below that would actually satisfy an independent auditor — not just what feels reassuring.

Did the system apply eligibility rules correctly?

Select evidence from the bank

Requirements with strong evidence

0 / 5

Auditability means a requirement is mapped to durable evidence: tests, logs, samples, and owners — not memory or a single reassuring number.

An audit-ready AI system preserves testable requirements, versioned documentation, representative decision samples, evaluation results, change history, access records, incidents, and remediation status. Auditors should be able to compare policy to behavior.

Independence matters. A team auditing its own launch may miss incentives, blind spots, or shortcuts that a separate reviewer would question.

Audit Questions

What requirement applies? What evidence proves it? Who reviewed the evidence? What failed? What changed because of the finding?

Checkpoint

Which artifact best supports auditability?