A risk manager’s guide to AI
Follow the full process from policy and evidence collection through analysis, controls, monitoring and communication.
Guides for risk managers who want to use AI in their day-to-day work and check the evidence behind each decision.
Follow the full process from policy and evidence collection through analysis, controls, monitoring and communication.
Trace changes to their source clauses, check who has authority and give the policy owner a short decision sheet.
Follow the process from scope to communication. Each lesson includes a worked example, a check for the risk owner and a copyable prompt.
Scope and authority
A risk assessment starts with the decision, the service at stake and the actions the system can take.
Identify and collect
Design the form, data checks and approval route before you ask AI to populate the register.
Qualitative assessment
Work one failure path through exposure, agent authority, control gaps, reach, harm and recovery.
Data science and assessment
Build a baseline, test it on later data and measure how much work an alert creates.
Disaster planning and scenario analysis
Convert a hurricane scenario into a continuity exercise with timed decisions, named owners and fallback routes.
Dependencies and global outages
Use the Cloudflare outage to test whether your fallback works when a shared service fails.
Quantitative assessment and simulation
Build a loss range, show the assumptions and calculate the chance of exceeding the budget.
Treatment and control testing
Compare control options by the harm they prevent, the work they add and the evidence you can collect.
Monitoring and review
Give every threshold a definition, an owner, a fallback and an escalation action.
Reporting and risk communication
Turn one checked finding into presentations, animations or requests for evidence that each role can act on.
Choose a failure path to test, then estimate what it could cost.
Qualitative assessment
Work one failure path through exposure, agent authority, control gaps, reach, harm and recovery.
Adrian Dunkley’s proposed method
Adapt Factor Analysis of Information Risk to measure how AI failures turn into losses at operating scale.