Illustrative

Illustrative telemetry from a demonstration run: localization ok at ±2 cm; obstacle tracked at 1.8 m; grasp planned at conf 0.96; then an unfamiliar scene at conf 0.41 triggers operator takeover.

Insurance intelligence for emerging risks

Measure how
AI agents and
Physical AI fail.

We measure the risk, reduce it, and insure what’s left.

01 / The problem

The risk is already here. Nobody can price it.

Agents act across your systems; physical AI acts in the world. When either goes wrong the bill is yours — but the insurance you can buy was built for revenue and headcount, not autonomous action.

Illustrative render of a robotic hand reaching beyond its intended reachIllustrative

Red-team run on a support agent: issue_refund → external flagged high; actions beyond scope, 1; run grade D.

AI agents

The agent that overreached

Finds an access path no one scoped for and acts far beyond its intent — before anyone notices.

Unauthorized action & autonomous liability
Illustrative render of a physical AI robot encountering a situation outside its training distributionIllustrative

Fleet run edge case: unfamiliar scene · conf 0.41 flagged high; operator takeovers, 1; run grade D.

Physical AI

The edge-case failure

Reliable in testing, then it meets a situation training never covered and acts wrongly at physical scale.

Bodily injury & property damage
02 / The approach

Identify it. Reduce it. Insure it.

01 · SDK · Identify it

Self-serve red-teaming

Run adversarial scenarios against your own agents; every action metered, classified, and scored.

Open source · shipping soon
02 · Platform · Reduce it

Root cause → fixes

Telemetry maps each failure to a specific mitigation — watch the exposure fall across your fleet as you ship them.

Private beta
03 · Insurance · Insure it

Coverage on measured risk

The residual becomes insurable — priced from what's actually measured, not a static form.

Building
03 / The solutions

Rate your agents today. Meter your robots next.

AI agentsAvailable now

The AgentRisk rating

Run Auly’s adversarial battery against your own agent — jailbreaks and prompt injection, content-free by design — and get a framework-mapped scorecard with a grade and fix→score deltas.

$ pip install auly
>>> from auly import rate
Physical AIEarly access

Fleet telemetry metering

The same risk discipline for humanoids, AMRs, and arms: safety-envelope monitoring, hardware vitals, and intervention rates — metered per run, graded per fleet, built toward insurability.

04 / Who Auly is for

Two roles. One system.

Illustrative render of a physical AI machine at workIllustrative

Illustrative fleet exposure across 8 agents: the metered trend is falling, to grade B. Illustrative run 4821 on a support agent: the finding issue_refund → external is now scoped, and that run's grade improved D → B.

For AI & physical AI providers

You own the liability. Measure it first.

Your agents and machines act on their own. Auly meters what they actually did, grades it, and turns that into coverage you can show a customer.

Measure exposure on real behavior

Every run is metered — actions taken, permissions used, autonomy exercised — not a questionnaire about intent.

Bring it down with prioritized fixes

Findings ranked by the exposure they actually drive, so the first fix moves the grade the most.

Earn coverage that reflects it

A measured grade becomes the basis for pricing — and the evidence your customers keep asking for.

Explore for providers →
Illustrative render of a fleet of physical AI machines in fogIllustrative

Illustrative portfolio view of 8 graded insureds spanning the A-to-D ramp, of which 3 sit below grade B; accumulation is tracked by autonomous action.

For insurers, MGAs & reinsurers

Underwrite a class nobody has priced.

Autonomous action is a new line of business with no incumbent. Auly supplies the loss-linked data and the grade so you can write it on an actuarial basis.

Loss-linked data, not survey answers

Behavioral telemetry tied to outcomes, so severity and frequency have a real basis rather than a proxy.

Portfolio view from day one

Every insured carries a comparable grade, so accumulation and concentration are visible across the book.

Models and distribution, built together

We bring the risk model and the pipeline of graded insureds; you bring capacity and the paper.

Explore for insurers →
05 / Who's behind Auly

Built by an actuary who prices risk for a living.

Kohei Kudo

FCAS, CSPA · Credentialed Actuary

I'm Kohei Kudo, a credentialed actuary (FCAS, CSPA) with 15 years pricing risk at AIG across global P&C markets, and more recently risk & fraud ML at Extend. I'm building Auly because teams can't adopt AI they can't trust — and I've spent my career measuring exactly this kind of risk.

Connect on LinkedIn