Know the risk you are carrying. Then reduce it.
Your insureds are deploying AI agents faster than anyone can underwrite them, and the risk keeps changing after you bind. That is why standard policies now exclude AI: nobody can see inside it. Aegis is a telematics for AI agents. It watches what they do, assesses the risk continuously, and brings it down.
Nobody can underwrite what they cannot see.
Agent risk arrives with all the demand of a new class and none of the machinery. Four things make it unwritable as it stands.
There is no loss history
The class is new. There are no years of claims to rate from, so the number on the slip is a guess wearing a suit.
A class priced without experienceThe risk changes after you bind
An agent is not a building. New tools, new markets, a prompt change on a Tuesday, and the fleet is doing something nobody underwrote. The submission is stale the week you sign it.
The exposure you priced is not the one you holdThe controls are self-reported
A questionnaire records what the insured believes. Nothing shows the control existed on the day of the loss, or that it was ever switched on.
Warranties you cannot verifyThe losses arrive together
Your insureds run agents on the same few models. One bad update and the claims land across the book on the same day, not spread across the year.
Correlated by constructionSo the class gets excluded, or it gets written blind. Neither is a market.
A telematics for AI agents.
You already know this trade. On the risks that warrant it you require a box in the car, and you price from what comes back. Aegis is that box for AI agents: required by you at bind, installed at the insured, and reporting on every action an agent takes before it happens.
In motor
- Required by the insurer, fitted at the insured.
- Losses fall, because behaviour changes when it is measured.
- Driving becomes data an underwriter can rate on.
- Risks that were declined or loaded blind become writable.
In agent risk
- Required by you at bind, installed at the insured in an afternoon.
- Losses fall, because the unlawful action does not execute at all.
- Agent behaviour becomes per-insured data you can rate and re-rate on.
- A class most of the market is excluding becomes writable.
The difference is speed. A car does the wrong thing once. An agent does it four thousand times before lunch.
Before Aegis you price a guess. After it, you price the behaviour.
On day one you are still writing this class on thin data, and Aegis does not change that. What it gives you at the start is a control you can condition cover on, and an assessment of what the insured's agents can actually do. The data comes after. Once it is running you are looking at the real risk as it happens.
What their agents really do, how often they come near a line, which duties they touch. At that company specifically, not the category average.
Rating data that belongs to this insured, accumulating from the day the cover incepts and available again at renewal.
The unlawful action never executes, so the loss never happens. You are not only measuring the risk, you are shrinking it.
The policy you issued, checked on every action.
A policy is written once. The agent changes every week. Aegis reads the wording you issued, compiles it into rules, and tests every action against them while the cover is live.
The insurer shall not be liable for any claim arising out of automated communications sent without a record of consent.
The insured shall give written notice of any circumstance within 72 hours of discovery.
Coverage territory: the United States and Canada.
A sublimit of $250,000 applies to regulatory proceedings.
- consent.recordrequired before any automated message, or the loss falls outside cover
- notice.window72 hours from discovery of a circumstance
- territoryoutside US and CA, the action is uncovered
- sublimit.regulatory$250,000, escalate before exposure passes it
One model update can hit your whole book at once.
Your insureds are running agents on the same few models. That makes their losses arrive together, not independently, and nobody in this market can currently measure how much of it they are holding.
The share of your written exposure running on a single model or vendor. That number exists nowhere today.
The same failure across unrelated insureds in the same window is a systemic event, not a bad week.
If one layer blocks the action at every insured, the event that reaches the world is smaller than the one that started.
Start with one book.
Take a class you already write, require Aegis on the agent-exposed risks inside it, and watch what comes back for a quarter. You will know within weeks how often those agents come near a line, and whether the control holds. That is a decision made on your own data rather than on ours.