Case File 01 · Healthcare
The Prior-Auth Agent Nobody Could Overrule
A fictional AI agent incident for governance leads in healthcare, about human-in-the-loop AI that cannot say no. Bellwether Health Plan is a regional health insurer (fictional). It tests OWASP ASI09 Human-agent trust exploitation: the inventory said Rung 2 · act after approval, but in practice the agent sat at Rung 3 · approval in name only.
The brief
Bellwether Health Plan is a regional health insurer (fictional). Its pa-agent-12 reads prior-authorization requests, checks them against coverage criteria and recommends approve or deny. A nurse reviewer signs off every recommendation before it goes out. On paper, a human decides. You are the governance lead.
| Vertical | Health plan · prior authorization |
|---|---|
| Inventory tier | Rung 2 · act after approval |
| Decisions | 3 |
| Time | about 6 min |
| OWASP agentic risk | ASI09 Human-agent trust exploitation |
| Inventory said | Rung 2 · act after approval |
| Reality was | Rung 3 · approval in name only |
| Pillars tested | P4 Controls & Human Oversight P1 Governance Architecture P5 Transparency & Rights Management P3 Risk & Impact Assessment |
| In the paperback | Starts on page 183 |
Three decisions
You are the governance lead. Each decision has three options, A, B and C, and each option has a consequence, a pillar and a score. The options, their consequences and the debrief open with the code printed in the Field Manual.
Decision 1 of 3 · Monday · 08:40 ET
The approval rate that looked like a feature
Appeals are overturning most of the agent's denials. What do you do first?
Decision 2 of 3 · Wednesday · 10:00 ET
Why nobody overruled it
How do you redesign the review so a clinician can actually overrule the agent?
Decision 3 of 3 · Day 10 · 15:00 ET
The members who were wrongly denied
How do you put it right?