AI oversight

What it is for

The EU AI Act expects meaningful human oversight of consequential automated decisions. This surface is where a person reviews what an agent advised.

How to reach it

Reports, in the footer of the navigation, then the AI oversight card. An individual decision has its own detail page.

Advisory, not automatic

A decision routed here is advisory: the agent proposes and a person disposes. That distinction is the point. An agent whose advice takes effect unreviewed is not being overseen, whatever the process document says.

What a reviewer sees

What the agent recommended, what it based that on, and which tools it used to get there. Enough to disagree with it, which is the only form of oversight that means anything.

How it differs from approvals

Approvals gate an action before it happens. Oversight reviews a judgement after it is made. An organisation subject to the AI Act generally needs both.

Where to go next

Record the feature these decisions belong to in Algorithm register.