← STACK/HUB · Compound Engineering / Agent Governance Series · RUBRIC v1.1

LEASH/14

Insurers and boards are starting to ask two questions about your agents: can they be contained, and can they be insured. This scores both, in 14 fields.

Why now: ISO/Verisk CG 40 47 and CG 40 48 (January 2026) are optional generative-AI CGL exclusion endorsements, not agent underwriting standards. CG 35 08 addresses products/completed operations; E&O, cyber and D&O wording is separate. Containment evidence may support a broker-reviewed renewal disclosure. Most teams have never scored either.
Scope check — 1 question

Do any of your AI agents take actions beyond generating content — spending money, sending messages, modifying systems, or triggering effects in the real world?

Reduced scope: containment isn't your exposure yet

LEASH/14 scores how readily agents that act can be stopped and contained. Agents that only read and generate content have little containment surface to score — running you through 14 containment fields would mostly measure questions that don't apply to you yet.

The trigger to come back: the day an agent of yours can spend, send, deploy, delete, or write to a production system. That's when insurability questions start having answers you'd need on file.

Until then, ATTEST/LOG (free) records what your agents are authorized to do — useful even for content-only agents: open ATTEST/LOG →

Insurability — 7 Fields

What an underwriter would ask before writing a policy

Containment — 7 Fields

Whether the agent can actually be stopped, and how fast
--LEASH/14 / 100
—

Field-by-Field Findings

Full Report

Underwriter-ready documentation + remediation
Renewal-disclosure draft explaining the CG 40 47/48 CGL exclusions, prioritized containment fixes, and documentation to review with your broker. Not an underwriting opinion.

Unlock Full Report — $89

One-time. Underwriter-ready narrative + prioritized containment fixes + documentation checklist.

Stripe checkout · instant delivery