An AI system sorts applications. An incorrect rejection can leave an eligible application unprocessed. Test whether a second professional review catches that error before dispatch, and prepare a control sheet for your own use case.
Work through case applications with known errors, count those found and missed, and check whether sufficient time is available for review. Adjust the control procedure or conditions of use and compare your result with the case explanation. For your own use case or the example, record risks, control steps, responsibilities and responses to deviations.
Take away your completed control sheet and error assessment. The exercise does not establish effectiveness in live operations. Binding risk acceptance requires a decision by authorised management and review of applicable requirements.
What you will work on
Assess AI risks for a specific use case using causes and consequences
Justify proposed acceptance boundaries and escalation triggers
Compare different controls against the same error sequence
Who it is for
Risk managers
Compliance managers
AI governance and control owners
Topics
Use boundaries and affected people
Risk scenarios and assessment uncertainty
Risk appetite, use boundaries and decision rights
Control effects and evidence requirements
Residual risk, monitoring and escalation
Your result
Your AI control sheet records what an error could cause, the limits of use, how to check the output and how to respond to deviations. It includes your assessment of the case applications and can be adapted to your use case.
Prerequisites
Basic risk management knowledge and a describable AI use case. Anonymised information is sufficient; a case study is planned.
Format and approach
Planned: 6–8 hours of self-study, including case decisions and self-checks with explanations.
AI governance needs clear decision rights for business owners, security, operations and other control functions. You jointly design approval and ongoing controls for an AI use case. Using a case, you practise reviewing results, handling deviations and repeating checks after changes.
Redesign a business process within your area of responsibility. First map the current workflow and identify steps that could be removed or simplified without AI. Then design an AI-supported alternative, specifying who decides, checks the results and handles exceptions.
Design a process for AI incidents, from reporting and limiting use through business review and a possible restart. Follow up decisions already influenced by an incorrect output.