The meter that makes AI risk insurable

AI liability insurance is real now — specialist programs write up to $25M per organization[F9] — but underwriting demands CONTINUOUS evidence customers don't have, and the majors are waiting for the instrument. We are the meter: the receipts already exist; ground truth is fed in as it arrives (complaints, spot checks — 5–25% is enough); a vigilant meter compares promise with outcome, built to look after EVERY new data point without false-alarming — the mathematically hard part, and the heart of our method.

Every planted calibration error caught

Every planted calibration error of the realistic kind was caught — five of five variants, 100% of runs — within a median of 6–53 confirmed outcomes. At a doubled error level with 10% follow-up: a median of ~150 outcomes to alarm, against a theoretical floor of 145. Faster cannot be built, for mathematical reasons.

The check-every-time method false-alarmed on up to 57% of clean runs; ours ran at 1.7–4.0% against a 5% budget.

Both sides of the policy

The company that wants to be insurable, and the insurer that needs the instrument to dare to write.

Status: figures measured in simulation on real model data; first partner pilot sought.

The Insurance Bridge, over time

Today

The meter: sparse ground truth in, promise-versus-outcome out, readable by your insurer.

Next

Partner pilots with specialist underwriters; the meter as the basis for backed warranties.

Long term

Continuous underwriting: premiums that track your measured risk instead of last year's questionnaire.

Looking for a first partner pilot

We're happy to walk an underwriting team through the meter in detail.