Insurance

Underwrite with AI.
Prove you did it fairly.

Insurance is the sector where AI decisions are already regulated by name. Colorado requires an attestation from the chief risk officer. New York requires a documented search for less discriminatory alternatives. The NAIC bulletin places AI governance inside existing market conduct law. Every one of those obligations is a question about what the model was given and why it decided what it decided — and that is a record-keeping problem before it is a modelling problem.

$10.36B

AI in insurance market in 2025, growing at a 35.7% CAGR

99%

of insurers have launched generative AI initiatives

separate US regimes already govern insurer AI by name

Dec 2027

EU AI Act high-risk obligations for life and health pricing

Request a Demo →See All Capabilities

The Problem

Adoption arrived first. The evidence requirements arrived second.

Insurers deployed AI across underwriting, rating, claims and fraud faster than they built the machinery to explain it. The regulators did not wait — and what they are asking for is provenance.

99%

of insurers have launched generative AI initiatives

76% have generative AI running in at least one business function, and 84% of health insurers now use AI or machine learning in operations. Adoption at that level means the governance question is no longer hypothetical for anyone in the market — it is a live examination exposure for nearly every carrier.

Industry generative AI adoption surveys, 2025–2026

2021

Colorado made insurer AI governance a statutory duty, not a guideline

SB 21-169, signed 6 July 2021, bars insurers from using an external consumer data source, algorithm or predictive model that produces unfair discrimination on protected characteristics. It requires a risk management framework, reporting to the Commissioner on methodology and ongoing monitoring, and an attestation from the insurer’s chief risk officer. The obligation is testable: prove you looked.

Colorado SB 21-169 / Division of Insurance rulemaking

68%

of organisations that suffered an AI incident had no AI governance policy

Up from 63% the year before. In insurance the cost of that gap is not reputational — it is a market conduct examination in which the carrier must reconstruct, after the fact, what data an automated decision consumed and why. Systems that keep no durable record of the decision cannot answer that question at any price.

AI incident and governance surveys, 2026

The Solution

How CLEOS addresses the insurer AI evidence problem

Three capabilities aimed at what the NAIC, Colorado and New York actually ask for — a record of what informed the decision, enforcement before the decision is made, and an audit trail that survives the session.

Decision Provenance for Underwriting and Rating

Every regime here converges on one demand: show what the model was given. CLEOS captures each tool call, data access and model interaction with full provenance, and keeps it in an append-only log rather than in the session that produced it. When an examiner asks which external consumer data source informed a specific rating decision eighteen months ago, the answer is a query, not an archaeology project. The same record supports the periodic monitoring Colorado requires and the quantitative testing New York specifies.

Colorado SB 21-169NY DFS CL No. 7Append-Only Audit Trail

Pre-Execution Enforcement of Eligibility and Rating Boundaries

Post-hoc review finds the violation after the consumer has been quoted. CLEOS enforces at the point of action: pre-execution hooks evaluate an AI action against policy before it executes and can refuse it. A prohibited data source, an out-of-appetite risk, or a decision that ought to require human sign-off is stopped rather than flagged. For carriers operating under an approved risk management framework, structural enforcement is materially easier to attest to than a review process.

Pre-Execution RefusalChief Risk Officer AttestationHuman Sign-Off Gates

Claims Automation That Keeps Its Reasoning

AI is cutting claims handling time and cost across the industry, with reported reductions of 20–30% in processing cost. The exposure is that speed without provenance produces decisions nobody can defend on appeal or in a bad-faith action. CLEOS gives claims AI persistent memory across the life of a claim — prior correspondence, prior determinations, prior adjuster reasoning — so the file compounds rather than resetting, and so a denial can be traced to what was known when it was made.

Claim-Lifetime MemoryDenial TraceabilityBad-Faith Defence

Why the Math Works

Where the insurance case is made

The insurance argument is not primarily a cost argument. It is that certain obligations cannot be met at all without a durable record of what the AI was given.

35.7%

CAGR of the AI in insurance market through 2034

From $10.36B in 2025 toward a projected $154B by 2034. Growth at that rate means governance infrastructure is being chosen now, at the point of deployment, rather than retrofitted later.

20–30%

Reported reduction in claims processing cost from AI

Industry-reported, across routine claims handling. CLEOS does not claim this figure — it is the prize carriers are already pursuing. What CLEOS addresses is the condition on keeping it: an automated determination that cannot be explained is a liability that grows with volume.

98.4%

Of context tokens eliminated per session

Independently recounted against a billing-grade tokenizer. For carriers running long-lived claims and underwriting files, the operative consequence is not the saving — it is that a file longer than the model’s context window remains workable at all.

Regulatory Alignment

The regimes that already name insurer AI

Insurance did not wait for a general-purpose AI statute. Three US regimes and one EU regulation already impose specific, documented obligations on automated insurance decisions.

NAIC Model Bulletin on the Use of AI by Insurers

Adopted 4 December 2023 and issued state by state. It sets out how insurance departments will read existing unfair trade practices, market conduct and corporate governance statutes when applied to AI used in underwriting, rating, claims, marketing and fraud detection. It expects a written AI systems programme and records sufficient to demonstrate it.

NAIC · Adopted 2023

Colorado SB 21-169

A statute, not guidance. Prohibits external consumer data sources, algorithms and predictive models that result in unfair discrimination; requires a risk management framework, reporting to the Commissioner on methodology and ongoing monitoring, and an attestation from the chief risk officer.

Colorado · Required

NY DFS Insurance Circular Letter No. 7

Adopted 11 July 2024. Sets expectations for AI and external consumer data in underwriting and pricing: analysis of unfair or unlawful discrimination, demonstration of actuarial validity, a governance framework, and documented searches for less discriminatory alternatives. It names the metrics — adverse impact ratio, denials odds ratios, marginal effects, standardised mean differences.

NY DFS · Required

EU AI Act — High-Risk Classification

Risk assessment and pricing in life and health insurance is classified as high-risk under Annex III, bringing logging, human oversight, technical documentation and post-market monitoring obligations. Under Regulation (EU) 2026/1744 the high-risk obligations apply from 2 December 2027 — later than originally legislated, but the documentation burden is unchanged.

EU · Effective Dec 2027

Buyers

Who in insurance buys CLEOS

Insurer AI decisions sit with the officers who personally attest to them. That is an unusually short path from problem to budget.

Chief Risk Officer

Signs the Colorado attestation personally. Needs the risk management framework to be demonstrable from records rather than described in a memo.

Chief Actuary

Owns demonstration of actuarial validity and the quantitative testing New York specifies. Needs the inputs to a model decision to be reconstructable long after the decision.

Chief Compliance Officer

Carries market conduct examination exposure across every state that issues the NAIC bulletin. Needs one evidentiary record that answers all of them.

Head of Underwriting

Accountable for decision quality and speed at once. Pre-execution enforcement keeps automated underwriting inside appetite without adding a manual review queue.

Underwrite with AI.Prove you did it fairly.