AI agents and insurance: what to hand underwriting

Insurers are already acting

CSIS reports that state insurance commissioners had “quietly approved more than 80 percent of carrier requests to exclude AI-related damages from corporate insurance policies,” citing a report by The Information from April 23, 2026.

Gartner predicts that by 2030 insurers, not regulators, will drive AI governance.

What the record shows

TrustGate returns a signed allow, deny or escalate decision before the action runs, for actions that go through it. The receipt binds the agent, action, target and exact arguments that were evaluated.

TrustGate reports 'could not check' separately from 'checked and found nothing'.

For an underwriter or broker, that is a record to read, not a policy document to take on trust.

5 questions worth asking an insured that runs agents

1. Which actions can the agent take?
2. Which of them need a person before they run?
3. What record exists that the agent did not write?
4. When a result cannot be confirmed, how is it reported?
5. Does the record separate 'could not check' from 'checked and found nothing'?

What this is not

• Not a coverage product, a rating or an insurability opinion.
• Not a promise of a lower premium.
• Not a lock. Signatures prove and detect. They do not prevent.

How an insured starts

Bring the insured, the agent they want to switch on, and the question underwriting keeps asking. We start with 1 workflow.