AI Liability Insurance: A Buyer's Guide

What AI liability insurance is, who needs it, and how the three forms of cover (endorsement, standalone, and performance guarantee) differ.

By Eric Ullrich, founder, Helix Risk Advisors. Facts checked November 14, 2026 against published carrier, MGA, and market sources.

The problem in one paragraph

Standard insurance forms were written before generative AI. Some carriers are affirming AI in their wording, some are adding endorsements, and some are filing exclusions. The same company can hold a Tech E&O policy that affirms AI, a GL policy with a new GenAI exclusion, and a D&O policy that is silent. Buying AI cover starts with knowing which of those you hold.

Builders versus deployers

Companies that build AI products face model performance, training data IP, and output liability. Companies that deploy AI face output errors, data disclosure, and reliance claims inside their own operations. The market sells different products to each: Armilla for builders, Testudo for deployers, and confusing the two is the most common placement error.

The three forms of cover

First, endorsements on Tech E&O or cyber (Embroker, Beazley, CFC, Corix for Vouch). Second, standalone AI liability policies (Armilla, Testudo, Relm, Mayflower). Third, performance guarantees (Munich Re aiSure), which pay when a model fails measured performance rather than when a claim arrives. Each answers a different question and they can be combined.

What it costs and what moves the price

There is no honest rate card yet. Underwriters price sector, autonomy, data sensitivity, governance, and limits. Programmes publish capacity, not prices: Armilla to $25M aggregate, Testudo around $9.25M to $10M (sources conflict), aiSure to $15M via Mosaic. Any site quoting you a firm AI premium without underwriting is guessing.

Frequently asked questions

Does our cyber policy cover AI?

Probably AI used to attack you, at the better markets. Liability for your AI's outputs is a different cover. Read the form, or send it to us and we will.

Do we need standalone AI cover at seed?

Usually not, if your Tech E&O affirms AI at full limits. Standalone becomes relevant when customers require dedicated limits, when the base form sublimits AI, or when you deploy agents with real autonomy.

Talk through your stack before you sign the next contract

Send us the term sheet, the customer MSA, or the renewal you are worried about. We will tell you plainly what it requires, what your current programme does about it, and what we would place differently.

Get Started   Take the coverage quiz