Risk · R-093 · Governance & Accountability
Weak independent assurance of AI
Second-line review and internal audit do not cover AI with sufficient skill or frequency, so control failures persist undetected.
- Severity
- Medium
- Audience
- Board · Heads-of
- Frameworks
- Clause 9 · GOVERN 4 · Principle 4
- Responsible AI domain
- Lifecycle stage
- Reviewed
- 2026-11-10
What it is
Assurance is the independent check that controls exist and work. The second line sets standards and challenges the first; internal audit gives the board an independent opinion. For AI both are often thin: the plan does not include it, the teams lack the skills, or reviews examine documents without testing systems. Validation of individual models is covered separately; this entry concerns assurance over AI governance as a whole.
How it happens
- The audit universe was drawn up before AI was in use.
- Reviewers rely on what the first line tells them.
- Assurance checks that a policy exists and stops there.
- Specialist skills are neither hired nor bought in.
Why it matters
The first line marks its own work. Weaknesses in inventory, tiering, oversight and monitoring go unreported, and the board's comfort rests on the absence of findings from reviews that never took place.
Warning signs
- No AI topic in the internal audit plan.
- Second-line reviews of AI produce no findings.
- Assurance staff have had no AI training.
- Audit testing stops at policy and does not reach a live system.
Mitigating controls
Primary controls
- Independent assurance and internal audit of AI Detective
- Places AI on the second-line and internal audit plans with tracked findings. Second-line review and internal audit cover AI governance and the highest-tier systems on a risk-based plan, test systems as well as documents, and track their findings to closure.
Supporting controls
- AI competence and resourcing plan Directive
- Gives assurance functions the skills to challenge. The skills, people, tooling and budget needed to build, validate, operate and oversee the organisation's AI are identified, compared with what exists, and the gaps are closed or the ambition reduced.
- Independent model validation Detective
- Independent validation is itself a second-line activity for models. A function independent of the developers tests each model's conceptual soundness, data, performance and limitations before approval and periodically afterwards, to a depth set by the model's tier.
- Speak-up route for AI concerns Detective
- Concerns raised by staff give assurance functions a source they would otherwise lack. Staff can raise concerns about how AI is being built or used, in confidence and without fear of reprisal, and each concern is assessed, investigated where warranted and answered.
Framework and clause references
| Framework | Clause | Title |
|---|---|---|
| ISO/IEC 42001:2023 | Clause 9 | Performance evaluation Internal audit and management review. |
| NIST AI Risk Management Framework (AI RMF 1.0) | GOVERN 4 | A culture that considers and communicates AI risk A culture of challenge. |
| SS1/23: Model Risk Management Principles for Banks | Principle 4 | Independent model validation Independent validation and review. |
Sources
- SS1/23: Model Risk Management Principles for Banks Bank of England PRA · official text
- ISO/IEC 42001:2023 – AI management systems ISO/IEC · framework
- NIST AI Risk Management Framework (AI RMF 1.0) NIST · framework
Related terms
How to cite
Axiom Verity (2026). Weak independent assurance of AI. Reference Library, R-093. https://axiom-verity.com/risks/weak-independent-assurance-of-ai. Reviewed 2026-11-10.
Found an error or a better source? Report a correction.