Skip to content

We score AI agents.
So an insurer can cover them.

A score solid enough for a risk carrier to underwrite against. Continuous assessment, remediation, pricing.

EPFL Blaze startup accelerator

Swiss product.
Data in Switzerland.

From EPFL's Blaze startup accelerator

How it works

Everything you need to deploy an agent, in one place.

Assess

Continuous scoring of your agents in production. The flaws before those who go looking for them.

Adversarial testingContinuous scoreDashboard
Remediate

Guided remediation. Your exposure falls continuously, not once a year.

Prioritised gapsScore trackingTimestamped evidence
Insure

Cover designed for agents. The score sets the premium.

Score-linked premiumDedicated limitPurpose-built covers
Garenzia Score The agent's score rises from 74 to 80 after remediation.
Annual premium · specimen CHF The annual premium falls from 9,600 to 8,400 francs.

Fix the flaws, the score rises, the premium falls.

Covers

Cover for what agents actually get wrong.

Unauthorised commitment

A discount, a deadline or a price promised without authority.

Third party

Processing error

An order, an invoice or a file handled wrongly.

Third party

Action outside scope

The agent does what it had no right to do.

Third party

Disclosure

Confidential information released by the agent.

Third party

Failure to escalate

A serious case handled alone instead of being escalated.

Third party

Crisis management costs

Investigation, legal counsel, notifications.

First-party loss

Indicative scope. Covers, limits and exclusions are set out in the policy.

Sovereignty

Your data never leaves your premises.

The assessment comes to your data. Your data does not come to us. The engine runs inside your environment; what comes out of it fits in a report.

No transfer to document, no processor to get approved, no copy to destroy afterwards. Your legal team has nothing to negotiate, because there is nothing to send out.

On your side

The engine runs on your infrastructure. Logs, prompts, client files: everything stays where it is.

No copy

Nothing is extracted, nothing is retained on our side, nothing trains a model. Neither during the assessment nor after it.

Read only

The connector observes. It never writes to your systems.

Insights

Read further.

What Europe's decided cases reveal about the price of an agent going wrong.

All insights

Get your first agent assessed.

Get an agent assessed