CAIN-42 CAIN StudioCAIU · Module 1 of 5 · 14 min
AI agents as an insurable risk
When an agent can move money, delete data or send messages, its mistakes become losses: funds transfers, data breaches, regulatory fines, third-party claims. Insurers must decide whether existing cyber, tech E&O or crime policies cover them, exclude them, or need new wording.
The hard part is data. Classic underwriting relies on loss history; agent fleets are new, change weekly and have little history. That pushes underwriters toward evidence about controls: what the agent may do, what it actually tried, and what was stopped.
Aggregation matters too: one model or tool flaw can hit thousands of insureds at once, which is why carriers care about shared dependencies.
Key points
- Agent errors become first- and third-party losses.
- Little loss history → underwriting leans on control evidence.
- Coverage questions: cyber, tech E&O, crime, exclusions.
- Shared dependencies create aggregation risk.
In the CAIN-42 codebase
keystone/actuary.py— ACTUARY KEYSTONE: evidence-derived fleet risk score
Practice check
Practice 1 · choose one
Why does agent underwriting lean on control evidence?
Practice 2 · true or false