Early access — measure your AI agent's risk

Identify your AI agents' risk.Mitigate it. Insure it.

Auly measures the risk in every action your agents take, helps you bring it down with targeted fixes, and insures whatever's left.

Your trust stack:SOC 2ISO 27001Insurance

01 — The problem

The risk is already here. Nobody can price it.

AI agents now act on their own — across your infrastructure, your customers, and your money. When one goes wrong, the damage is real and the bill lands on you. But the insurance you can buy was built for revenue and headcount, not autonomous action.

It has already happened — to companies of every size

Air CanadaFeb 2024

Its support chatbot promised a retroactive refund option that its own policy explicitly excluded. A tribunal ruled the airline liable for what its AI said.

Held legally liable

Replit / SaaStrJul 2025

A coding agent deleted a live production database during a code freeze — a database of 1,200+ executive and company records — then generated fake data to conceal it.

Production data destroyed

PocketOS / CursorApr 2026

A Cursor coding agent discovered an API token with unrestricted permissions and deleted PocketOS's entire production database in 9 seconds, along with all volume-level backups.

Production data destroyed

Carriers can't price what they can't see.

  • No claims history for autonomous agents
  • No actuarial model for agent behavior
  • No policy built for how agents act

So they decline the risk — or quote a number they invented.

You're priced on the wrong numbers.

Sets your premium

  • Annual revenue
  • Headcount
  • A one-time form

Sets your real risk

  • Agent actions
  • Permissions
  • Autonomy

What you're charged for and what can actually hurt you don't match.

02 — How it works

We sit in the path — capturing risk, reducing it, and insuring what's left.

Auly lives in your AI agents' execution path. Every action is metered, classified, and scored as it happens; we help you bring the risk down, then cover the residual — priced from what we actually measure.

Continuous monitoring · live
Step 1 — In the path: capture & classify
AI agent
acts
meter · classify · score
Auly
always on
prod-db
payments
repo
email
CAPTURE

Meter every action

Tool, argument schema, counterparty — in the path, no rebuild.

CLASSIFY

Tag the risk

A semantic risk class per action — the real exposure base, not raw logs.

SCORE

Detect & alert

Authority breaches, anomalies, override spikes — caught the instant they happen.

Step 2 — The loop: evaluate → mitigate → insure, on repeat
runs continuously · re-rated as your risk changes
01 · EVALUATE

Score the risk

A risk-weighted grade, every factor traceable to a failure mode.

B-
Reach
Scoping
Rollback
02 · MITIGATE

Reduce it

Prioritized fixes; watch the exposure fall in real time.

CA-
Least-privilege tokens
Approval gate on deletes
Isolate backups
03 · INSURE

Cover what's left

Coverage priced from your measured risk — not a static form.

● INSURABLE · A− (illustrative)
$1,900/mo
example · priced from telemetry

03 — Why Auly

They scan the perimeter. We live in the path.

IN-PATH

We see the behavior

Active-cyber insurers scan exposed assets from outside. Auly runs where the agent acts — the one place that behavior is visible.

CONTINUOUS

We re-rate as it changes

Not a point-in-time questionnaire. The score and the price move with the agent's real, measured risk.

THE MOAT

Loss-linked telemetry

Every factor is logged at scoring time, so heuristics become fitted weights as incidents accrue — a dataset no one else holds.

See the risk in what your agents do.

Join the early-access waitlist — get your agents scored as we roll out.

We'll only use your email for early-access updates. See our Privacy Policy.

Prefer to chat? support@kalta.ai