The method
Structured analytic techniques, applied to AI risk
Each exercise is built on named, defensible methods from the intelligence field, not improvisation. The facilitator probes the team's reasoning rather than running a discussion.
The four steps
Pre-work
A scoping call and a short questionnaire establish your sector, your AI footprint, and the decisions that would matter most under stress. The scenario is then built for your situation.
Facilitated session
Your team works through a realistic AI-failure scenario in timed decision rounds. New information arrives as injects, the way it would in a real incident.
Decision stress-test
A Key Assumptions Check and competing scenarios push the team's reasoning until it bends, exposing the assumptions no one had stated out loud.
After-action report
A written record: the findings that matter, specific and owned recommendations, and the early-warning indicators that signal the scenario is starting to occur for real.
The techniques
What makes it intelligence-grade
The same methods used to assess real threats under pressure and incomplete information are what a tabletop exercise simulates. Three carry the weight of the session.
- Key Assumptions Check. The team's unspoken assumptions about its AI systems are written down and challenged one by one.
- Multiple scenario generation. The exercise tests more than one way the failure could unfold, so the response is not tuned to a single guess.
- Indicator validation. Each session ends with the observable signals that a risk is becoming real — the part most competitors leave out.
What you keep
The after-action report
The durable output your team circulates, and your board or auditor can read.
Executive summary
The three findings that matter most, in plain language a director can read in two minutes.
Findings
The gaps in the decision chain, the unchallenged assumptions, the missing roles or authorities.
Recommendations
Specific, owned, and time-bound. This is where the value lands.
What happened
The decisions the team made, where it hesitated, what it missed.
Scenario & participants
What was tested and who took part, so the record stands on its own.
Indicators to watch
The early-warning signs that this scenario is starting to occur for real.
See whether the method fits your risk
A short call to talk through your AI footprint and which scenario would test it hardest.