What entered?
Were the right inputs and evidence considered?
An AI workflow can satisfy individual control checks while a consequential failure passes through the joins. Assurance must look at the whole chain.
Were the right inputs and evidence considered?
How did model outputs and controls interact?
Was oversight meaningful or merely procedural?
Which risks, unknowns and dependencies still matter?
Research identifies mechanisms worth investigating. Actual assurance conclusions require evidence from the organisation's own workflow.