The AI Assurance Readiness Checklist
Checklist
Check whether your evidence would stand up to an independent AI assurance review before you commission one.
The claim to assure
- Write the claim in testable terms, with a measure and a threshold.
- Name who relies on the claim: customers, regulator, board, users.
- Record which standard or framework the claim refers to, if any: ISO/IEC 42001, ISO/IEC 42005, NIST AI RMF.
Evidence available
- A current system description and model or system card.
- A risk and impact assessment with owners and dates.
- Test results, including results broken down by relevant user groups.
- Monitoring records and incident logs since launch.
- Decision records showing who approved release and changes.
Access the assessor needs
- Confirm what access you can give to the model, data, logs and people.
- Identify contractual limits from model or data suppliers that restrict access.
- Agree confidentiality terms before sharing technical material.
Scope limits
- State what the review will not cover.
- Agree how long findings remain valid if the system is retrained or updated.
- Ask how the assessor's people are qualified and what method they use.
Start a conversation
Have a product or AI decision to make?
Useful first calls usually start with one unclear decision, a deadline and a team that needs a practical next move.
Tell us about it