Risk and reviews

Screen every AI system, rate the ones that matter, and record each review and approval. Effort goes where the risk is.

Claims triage agent · risk screeningSet against your risk appetite
Data sensitivity4
Autonomy3
External actions4
Financial impact3
Customer impact2
Critical system access2
Tier 3Rule: any dimension at 4 or above sets Tier 3 or higher.Reviewed quarterlyAccountable: Head of Claims
Six dimensions, a written rule and a tier anyone can follow. Illustrative scores.

Screening first#

A short set of questions sorts systems quickly: does it use personal information, does it make or substantially help make decisions that significantly affect people, does it support a critical operation, can it act on its own. The answers are recorded, a rule suggests a rating, and a person confirms it.

Risk dimensions#

DimensionWhat it asks
Data sensitivityDoes it touch personal, financial or health information?
AutonomyDoes it recommend, decide with review, or act on its own?
External actionsCan it send, pay, change records or contact customers?
Financial impactWhat is the cost of a wrong output or action?
Customer impactCould a customer be treated unfairly or harmed?
Critical system accessDoes it reach systems that support critical operations?

Reviews and approvals#

A review records who checked what, when, and the outcome: a pre-deployment assessment, a periodic review, a security assessment or a change review. Approval to go live sits in the review, apart from the system's governance state, so "may we go live?" has a clear answer.

Obligations that apply#

From the control registry, each system shows the frameworks and obligations relevant to it, based on what it is and what it does. See Controls and obligations.

Automated decisions and the Privacy Act#

Screening flags systems that use personal information to make, or substantially help make, decisions that could significantly affect people. That supports your own assessment for the privacy policy changes that start on 10 December 2026. CognaOne helps you find and describe these systems; it does not certify compliance. See Privacy Act: automated decisions.

Reassess when things change#

When a release changes what a system can do, the assistant reads the change, proposes new answers with its sources and lists the controls that would follow. Ratings change only when a person confirms.

Last reviewed October 2026Suggest a change