AI risk register template

AI inventory versus risk register

An AI inventory records what tools and use cases exist. A risk register goes further by describing what could go wrong, who owns the risk, what controls are in place, what residual risk remains, and what decision the business made. Small teams often need both, but the first risk register should stay practical enough to maintain.

What an AI risk register means

An AI risk register is the operating record that connects an AI use case to its business risk, controls, owner, and review cadence. It is not just a list of tools. The register should show why a use case is low, medium, or high risk; what mitigation exists; what residual risk remains; and who accepted or escalated that risk.

AI risk register template download fields

A useful template download should include fields that let a team move from discussion to action. At minimum, include use case, system or vendor, owner, department, user group, data category, output use, decision impact, risk statement, likelihood, impact, controls, residual risk, mitigation owner, due date, approval status, and review date.

• Use case and AI system: what the workflow does and which tool supports it.
• Data and output: what goes into the system and how generated output is used.
• Risk and controls: what could go wrong and which safeguards reduce the risk.
• Decision record: approve, restrict, escalate, block, or review later.

Risk statements and owners

Each entry should describe a concrete risk, not just a tool name. A useful statement identifies the use case, affected data or people, possible harm, owner, existing controls, and escalation path. Ownership matters because risks without owners rarely get reviewed after launch.

Inherent risk, controls, and residual risk

The register should separate risk before controls from risk after controls. Inherent risk can consider data exposure, decision impact, external visibility, vendor dependency, and affected people. Controls can include approved-tool limits, human review, vendor opt-outs, data minimization, logging, training, and management approval. Residual risk is what remains after those controls are applied.

Human oversight and impacted people

AI risks become more serious when outputs influence employees, customers, applicants, patients, students, borrowers, or other people in important decisions. Record whether a qualified person reviews the output, whether the person can challenge the result, and whether the workflow should be escalated before use.

Review cadence and acceptance

Set review dates based on risk, not convenience. Low-risk productivity use can be reviewed periodically. Sensitive-data, customer-facing, vendor-dependent, or high-impact uses need tighter review. When a risk is accepted, the register should show who accepted it, why, for how long, and what evidence supported the decision.