Control · C-087 · Third Party & Supply Chain
Embedded AI feature review
Supplier release notes, terms and product settings are monitored for new AI features in products already in use, and each feature is assessed and approved before it is enabled.
- Control type
- Detective
- Risks addressed
- 3
- Audience
- Heads-of · Practitioner
- Frameworks
- GOVERN 6 · Principle 1
- Responsible AI domain
- Lifecycle stage
- Reviewed
- 2026-11-11
What it does
It catches AI that arrives inside software the organisation already owns. Someone is responsible, for each significant product, for watching what the supplier releases. A new AI feature is held off by default, routed through intake like any other proposed use, and switched on only when approved. The owner of the supplier relationship is accountable; IT administrators control the settings. It operates continuously, at each supplier release.
How to implement it
- Assign an owner for release monitoring to each significant product.
- Set products to keep new features off until enabled by an administrator, where the product allows.
- Review release notes and updated terms for AI features, data use and new sub-processors.
- Send each new feature through use-case intake, tiering and data protection screening.
- Record enabled features in the AI inventory.
- Periodically audit product settings against what was approved.
Evidence it operates
Design: a procedure naming product owners and the default position on new features. Operation: release review records; intake records for embedded features; inventory entries; settings audits showing only approved features enabled.
Where it fits
In operation, as the check on a route that bypasses procurement. It is the primary control for unassessed AI in third-party products and supports incomplete AI inventory. It feeds AI use-case intake and approval and the AI inventory.
Limits
Suppliers do not always announce AI features clearly, and some cannot be switched off. The number of products and the pace of releases make complete coverage impractical, so effort has to be directed at products that hold sensitive data or support important decisions. Features enabled for individual users may never pass an administrator.
Risks this mitigates
Primary mappings
- Unassessed AI in third-party products High
- Monitors existing products for new AI features and assesses them before they are enabled. A supplier switches on AI features inside a product already in use, bypassing the organisation's assessment and approval.
Supporting mappings
- Incomplete AI inventory High
- Finds AI switched on inside existing products. The organisation cannot list the AI systems and models it builds, buys or uses, so risk assessment, oversight and regulatory duties are applied to only part of what it has.
- Workplace and education emotion recognition High
- Finds emotion-inference features switched on inside workplace products. An AI system infers the emotions of people at work or in education, for a purpose other than medical or safety reasons.
Framework and clause references
| Framework | Clause | Title |
|---|---|---|
| NIST AI Risk Management Framework (AI RMF 1.0) | GOVERN 6 | Third-party and supply chain AI risk Third-party AI in scope. |
| SS1/23: Model Risk Management Principles for Banks | Principle 1 | Model identification and model risk classification Vendor models in the inventory. |
How to cite
Axiom Verity (2026). Embedded AI feature review. Reference Library, C-087. https://axiom-verity.com/controls/embedded-ai-feature-review. Reviewed 2026-11-11.
Found an error or a better source? Report a correction.