Insurance for AI startups, before the pilot becomes an MSA.
The buying problem is not a generic tech package. It is model output, training data, and a customer who will ask who is liable when the answer is wrong. We place cyber and technology E&O with wording that contemplates that, then watch for the round that triggers D&O.
What underwriters want to know
How you build: train your own models, fine-tune, or orchestrate third-party APIs. Whether outputs are assistive or used for credit, hiring, health, or legal decisions. What data you train on, and whether you have rights to it. Those answers change E&O wording more than your logo does.
We are not a carrier and we do not bind online. A licensed advisor reads the file. We aim to respond within one business day. This page is not a finding that any coverage applies to your model.
The claim that is actually new
- A customer relied on generated output and says the loss is yours
- Training or customer data leaked from a pipeline or a prompt log
- A provenance or IP dispute over data used to train or fine-tune
- A pilot converted to an MSA with indemnity you did not price
Standard cyber still matters. It is not the whole file.
Today
As you grow
Questions we hear
Sometimes, if the carrier treats you as software. Sometimes the form is silent on model output, or it excludes it. That silence is the problem. We ask how you build and how outputs are used, then pick markets that will put it in writing.
If you already have a pilot, a data set, or a customer asking for a certificate, it is not too early. If you are still in a notebook with no users, we can wait. The form lets you say pre-revenue.
No. Insurance is not legal advice and not a license. We place coverage subject to underwriting and policy terms. Counsel owns the product and data-rights questions.
Related guides
Tell us how the model is used.
The application opens on AI / machine learning so the file matches the product.