AI system intake SOP

Purpose

Give every AI system a documented risk tier and a recorded set of controls before it goes into use, so risk is identified and owned at intake rather than discovered after the fact. Legal content last reviewed: 2026-09.

Scope

Applies to any AI system a team proposes to introduce, build, buy, or extend to a new use — including a new use of a system already on the inventory.

Roles

• Requesting team — describes the use case, the data it uses, and whether a person reviews each output.
• AI governance lead — runs the intake, assigns the tier, and maintains the AI inventory.
• Legal — reviews any prohibited-practice concern and any high-risk tiering before approval.
• Security — reviews the supplier and data-handling controls for the system.
• Data protection — confirms personal-data use and provenance, and any data-subject impact.

Frequency

Every new system at proposal; every material change of use; a full re-review every year.

Steps

1. Restate the use case: purpose, affected people, decisions made or supported, and whether a person reviews each output XW-021
2. Record the data used, whether it includes personal data, and its provenance XW-028
3. Check prohibited practices first; if one might apply, stop and escalate to legal before anything else XW-021
4. Check the high-risk areas (employment, essential services, education, law enforcement, critical infrastructure) and tier accordingly XW-021
5. Otherwise tier as limited-risk or minimal-risk XW-021
6. List the required controls for the tier, mapped through the crosswalk and the NIST AI RMF functions XW-024
7. Identify the supplier and model provider and assess them XW-023
8. Add the system to the AI inventory using AI inventory record template XW-022
9. Enter any identified risk in the risk register with an owner XW-024
10. Record the approval decision and the approver, with legal review for any high-risk or prohibited result XW-020
11. Set a re-review date, and re-run intake on any change of use XW-021

Evidence to retain

• The restated use case, including data used, personal data, and provenance.
• The risk tier assigned and the reasoning behind it.
• The list of required controls for that tier.
• The supplier and model-provider assessment.
• The AI inventory record.
• The approval decision, the approver, and any legal review carried out.
• The re-review date set.

Related

• Governance framework crosswalk (crosswalk) — supplies the framework rows this intake's controls and tier map to.
• ITAM maturity check (itam-maturity) — its inventory and lifecycle questions extend to AI systems entering the estate through this intake.
• Application rationalization (rationalization) — takes over an AI system once it's on the inventory, to decide whether it's still worth keeping.
• Access review pack (access-review) — reviews who can use or administer an AI system once this intake has approved it.
• Stand up a governance program (program-setup) — stands up the sponsor and AI governance lead role this intake reports its decisions to.