Glossary
What is an AI audit?
An AI audit is an independent check of how a company uses and controls AI. The auditor compares what the company says it does, in its policy and its framework, with evidence of what really happens. The evidence is what the audit turns on.
By the AIBound team · Updated 17 November 2026
Key takeaways
- An AI audit compares the company's AI rules with evidence of what happens in practice.
- Auditors ask for three things: the list of AI in use, the risk judgment on each and the decisions made.
- Evidence built by hand is out of date when the auditor reads it, so a live record matters.
Why an AI audit matters
A company can say it controls AI. An audit asks it to prove it. Customers, regulators and boards rely on that proof because it comes from someone outside the team that runs the program.
Most teams find the hard part is the evidence. The auditor asks which AI employees use, and the answer sits in a spreadsheet built from a survey. The agents, MCP servers and skills nobody declared are the ones missing from it.
How an AI audit works
Audits differ by auditor and by the framework used. Most follow four steps.
- 1 Scope. Agree what the audit covers: which teams, which AI and which rules or framework.
- 2 Evidence. The company hands over its policy, its list of AI in use, its risk judgments and its record of decisions.
- 3 Testing. The auditor checks samples to see whether the record matches what is really in use.
- 4 Findings. The auditor reports what works, what falls short and what to fix by when.
AI audit vs AI risk assessment
A risk assessment is the company judging its own AI. An audit is someone independent checking that the judging, and the control that follows, really happen.
| Question | AI audit | AI risk assessment |
|---|---|---|
| Who does it? | An independent auditor, inside or outside the company | The team that owns the AI or the risk |
| What does it look at? | Whether the rules are followed, with evidence | How risky each AI tool or use is |
| What does it produce? | Findings and actions to fix them | A risk rating and a decision on each tool |
Sources
- NIST glossary: audit · csrc.nist.gov
Sources checked 19 November 2026
Examples
AI audit examples
FAQ
Frequently asked questions
What is an AI audit?
An AI audit is an independent check of how a company uses and controls AI. The auditor compares the company's rules with evidence of what happens in practice: which AI is in use, how each tool was judged and what was decided. The result is a set of findings and fixes.
What evidence does an AI auditor ask for?
Auditors usually ask for the AI policy, the list of AI in use, the risk assessment for each tool and the record of decisions, such as what was allowed or blocked and by whom. They also ask how the list is kept current, since a survey is out of date quickly.
Who carries out an AI audit?
It depends on the purpose. An internal audit team checks the company's own program. A customer or regulator may send an outside auditor. A certification body audits a company that wants a certificate against a standard. In each case the auditor is independent of the team that runs the AI program.
How do you prepare for an AI audit?
Start with the list of AI employees really run, taken from their devices and not from a survey. Record how each tool was judged and what was decided. Keep the record as the work happens. When the auditor asks, the answer is then an export and not a new project.