What does AI insurance cost for your agent? Check the tool permissions your agent actually has — email, code execution, payments — and get an illustrative premium range with the reasoning, grounded in real 2026 carrier signals.
Check everything your agent can do without asking you first.
The weights above are a fixed, transparent teaching table — they reflect which permissions create the loss scenarios carriers actually write about. They are not actuarial.
By counting what the agent can touch. Underwriters model injection risk from tool permissions: email and code execution multiply exposure, payment access multiplies it again. This calculator walks through your agent’s permissions and estimates the premium band carriers would assign. The cheapest policy is reducing permissions — every tool you remove is risk you do not have to insure.
There is no public rate card for AI agent liability insurance. Standalone capacity exists — Armilla AI, a Lloyd's coverholder, raised its AI liability limits to $25M per organization in January 2026 — but premiums are underwritten per deployment. This calculator gives an illustrative range from a fixed scoring table: your agent's tool permissions drive a risk score, and the score maps to a band. It is educational, not a quote.
Standalone AI liability policies, such as Armilla AI's, are built for how AI systems actually fail: AI model errors, harmful or misleading outputs, AI agent failures including incorrect decisions and improper tool use, data leakage via outputs, AI-driven property damage, and defense costs for AI regulatory investigations such as under the EU AI Act.
If your agent takes real-world actions — sending email, moving money, writing code or databases — you carry real-world liability. Since January 1, 2026, ISO's generative AI exclusion endorsements (CG 40 47, CG 40 48, CG 35 08) let carriers remove AI-related claims from standard commercial general liability policies, so your existing policy may no longer respond to agent-caused losses. Check your renewal paperwork.
In January 2026, ISO (Verisk's standard-forms division) introduced three optional endorsements for commercial general liability policies: CG 40 47 (broad generative AI exclusion), CG 40 48 (advertising injury only), and CG 35 08 (products/completed operations). When attached, they remove coverage for claims arising from generative AI — closing the old "silent AI" gap where coverage was ambiguous.
Each tool permission carries a fixed weight (code execution and payments highest, read-only access lowest), plus a surcharge if no human approves irreversible actions. The total becomes a 0–100 risk score and a grade A–F. The illustrative annual premium is the score times $120, banded plus or minus 25%. The table is fixed and transparent — not actuarial, not a quote.