Assurance methods
Different questions need different methods. Several methods can support the same decision.
Framework assessment
A structured self-assessment against a whole framework, reviewed and approved. Example: an EU AI Act self-assessment for a project.
Control Test
A check that one control operates. Record the result, notes and evidence reviewed. Example: sample 20 recent decisions and confirm each had the required human review.
Assessment workspaces
Structured reviews for one system, such as a DPIA, bias and fairness assessment, or vendor AI risk assessment.
Technical Evaluation
A measured result against a threshold. Example: the required accuracy is 94% and the observed accuracy is 91%, so the gate fails.
Human oversight review
A logged review by a person: scheduled, triggered by an event, ad hoc, or an override of the system's output.
Last reviewed 2 October 2026
Next stepTechnical evaluationsA Technical Evaluation measures AI behaviour against a defined method, dataset and threshold. You run it in your own environment and submit signed results to AIGovernr.