
Insurance
Insurance AI Readiness Guide
Insurance is the business of telling risks apart. AI makes that faster and sharper, and puts it directly on the decisions regulators and courts care about most: who gets cover, at what price, and whether a claim is paid.
By Kesho Partners
9 minute read
Insurers adopted AI early and widely, because it fits the work. But the systems now sit on the three decisions that define an insurer's relationship with its customers: underwriting, pricing and claims. Supervisors say existing insurance law already covers them. Courts are starting to ask insurers to explain how their systems work. An insurer that cannot produce that explanation on request is not ready, however good its models are.
The most AI-intensive part of financial services
In the Bank of England and FCA's 2024 survey, 95% of UK insurers used AI, the highest share of any financial sector. US regulator surveys found 92% of health insurers using, planning or exploring AI, with motor at 88% and home at 70%.
Controls have not kept pace. Nearly a third of US health insurers said they do not regularly test their models for bias or discrimination. In Kenya, the regulator estimates about a quarter of insurance claims are fraudulent, which explains the appetite for AI in claims, and the stakes when it goes wrong.
Price: when differentiation becomes discrimination
Insurers are allowed, and expected, to treat different risks differently. The law limits which differences they may use. EU insurers have not been able to use sex as a rating factor since the Test-Achats ruling took effect in 2012.
AI makes that line harder to hold. A model can rediscover a prohibited factor through proxies such as postcode, occupation or device type without anyone intending it. External consumer data, which the customer never supplied, adds both predictive power and legal exposure. Colorado already requires life, motor and health insurers to test external data and models for unfair discrimination. Under the EU AI Act, AI used to assess risk and price life and health cover for individuals is high-risk.
In plain terms: an insurer must be able to show not only that a model predicts well, but that what it uses is lawful.
Claims: automating the decision not to pay
A policy is a promise to pay. Automating the decision not to pay is where the sharpest litigation has landed.
In the United States, a class action alleges that UnitedHealth used an algorithm to cut off post-acute care for Medicare Advantage patients, and that most appealed denials were overturned while very few members appealed. UnitedHealth says the suit has no merit. In March 2026 a federal court ordered the company to produce internal documents about the tool.
A separate case against Cigna over a claims-review tool was allowed to proceed in 2025. Cigna says the tool is not AI.
Neither case has been decided. Both show that a court can require an insurer to explain how a claims system works, and that "a person reviewed it" will be tested against how much time and authority that person really had. We set out where to automate and where people must stay in charge in our claims article.
What supervisors now expect
Insurance supervisors have converged. Almost all say current rules on governance, conduct and unfair discrimination already apply to AI.
| Supervisor | Approach | What it means |
|---|---|---|
| IAIS (global) | Application paper, July 2025, applying existing core principles to AI | The global baseline: proportionality, governance, robustness, explainability, fairness and redress. |
| EIOPA (EU) | Opinion on AI governance, August 2025 | No new rules; explains how existing law applies to every AI use. |
| NAIC (US) | Model bulletin adopted by 29 jurisdictions; examination tool piloted in 2026 | Expect a written AI programme to be examined. |
| FCA (UK) | No new AI rules; Consumer Duty and pricing rules apply | Fair value must be evidenced whatever sets the price. |
| IRA (Kenya) | Drafting rules on cyber security, data protection and digital insurance | Data protection law binds today, including on solely automated decisions. |
The US examination questions are public. Any insurer, anywhere, can use them as a dry run: what AI is used, how it is governed, which models are high-risk, and what data feeds them.
Five questions every insurance board should ask
- Show me every AI system we use, with an owner and a risk rating. Include vendor models and external data.
- For our most material pricing or underwriting model, who validated it, and what did they find?
- When a claim is declined with a model's help, who reviews it, and how often do they disagree? A reviewer who never disagrees is not reviewing.
- Have we tested pricing and claims outcomes for unfair discrimination, including through proxies?
- Which external data and vendor models feed decisions about policyholders, and what do the contracts let us see?
Where to start
Start with an inventory that includes vendor models and bought-in data. Rate each system by its effect on customers, then test the highest-rated for fairness and explainability. Check that human review of declined claims is real, measured and recorded. Take the results to the board as decisions with owners and dates.
Our guide covers all ten areas: underwriting and pricing models, claims and fraud, external data, vendors, security and people, with the evidence to request, red flags, 42 questions and a 100-day plan.
The Insurance Executive's Guide to AI Readiness and Governance
Related reading
Sources
- Bank of England and FCA. Artificial intelligence in UK financial services, 2024.
- NAIC. Artificial intelligence, survey results 2022–2025.
- Fenwick. Tracking the evolution of AI insurance regulation.
- The Standard. AI advantage: insurance innovation in Kenya.
- Colorado Division of Insurance. Protecting consumers from unfair discrimination in insurance practices.
- EIOPA. Opinion on AI governance and risk management, August 2025.
- Forbes. The algorithm that counted on no one appealing, June 2026.
- Fenwick. As above, on the Cigna case.
- IAIS. Application paper on the supervision of AI, July 2025.
- FCA. AI approach.
- NAIC adoption map. As reported, October 2026.
- Beinsure. Kenya's IRA drafts digital insurance rules.
This article is general information, not legal advice.
Related service
Financial services
Independent AI readiness and governance assessments for insurers, covering pricing, underwriting and claims.
Explore the service