AI Risk & Insurance

AIRM-Core

AI risk modeling and telemetry underwriting engine

Overview

AIRM-Core screens how well a company controls the risks of its AI systems and turns that evidence into a risk score an insurance carrier can use for underwriting. It reads a company's architecture documents, collects operational telemetry, and shares only the final scores, never the company's private evidence.

The challenge

Insurers want proof that an AI system is well governed before they cover it, but companies can't hand over their architecture, logs and cloud credentials to a third party.

Key features

Multi-format architecture intake

Reads system specs and diagrams in Word, Excel, CSV, draw.io, Mermaid, PDF, Markdown, JSON and YAML.

Automatic telemetry planning

Works out which telemetry packages a system needs and lets engineers steer it with directives.

Signed telemetry packages

Every package is Ed25519-signed by the carrier and verified on arrival. Anything unsigned or mismatched is refused.

Local metric collection

Pulls metrics from sources such as AWS CloudWatch and Datadog, with credentials kept only in the container's memory.

Score-only export

Proprietary evidence stays inside the company; only allow-listed risk scores are sent to the carrier.

Hardened local portal

Browser UI with a strict content security policy, same-origin checks and no third-party scripts.

How it works

The company uploads its architecture documents to the assessment container running in its own environment.

The container requests the telemetry packages it needs and verifies the carrier's signature on each one.

Local collectors gather operational metrics and the engine scores technical and governance controls.

Only the combined, technical and governance scores are sent to the carrier's receiver for underwriting.

Technology

PythonGoDocker ComposeEd25519 signingmTLSAWS CloudWatchDatadog

Chatbot Compare MyCommunitySales