AI in Insurance: Claims, Underwriting and Fraud Detection

Insurance has always been a data business. Insurers price risk using data, decide claims on evidence and fight fraud by spotting patterns. So it is no surprise that AI has spread into almost every step of how policies are sold, priced, serviced and paid.

What has changed is the scale and the scrutiny. AI now reads claim photos, drafts adjuster notes and scores applications in seconds. At the same time, state insurance regulators have made it clear that “the algorithm did it” is not a defense. This guide walks through where AI shows up across the policy lifecycle, what it does well, where it goes wrong and what regulators are watching.

Where AI fits in the insurance policy lifecycle

AI is not one tool. It is a set of techniques, from predictive models to computer vision to generative AI, applied at different points:

The pattern across every row: AI helps most when it speeds up routine work and routes hard cases to people, and it causes the most trouble when it quietly makes consequential decisions nobody can explain.

AI in underwriting and pricing

Underwriting is where insurers decide whether to offer coverage and at what price. AI helps in 3 main ways:

• Data prefill. Instead of asking applicants dozens of questions, systems pull property, vehicle or business data from third-party sources and fill the application automatically.
• Risk scoring and triage. Models score applications so simple, low-risk ones can be approved quickly and complex ones go to an experienced underwriter.
• New data sources. Telematics from cars, aerial and satellite imagery of roofs and property, and connected sensors give underwriters more detail than traditional forms.

In commercial lines, generative AI is increasingly used to read broker submissions, loss runs and engineering reports and summarize them for underwriters. That is essentially intelligent document processing applied to insurance paperwork.

The concern regulators raise is proxy discrimination. A model may never use race or another protected trait directly, but it can rely on data that closely tracks it, such as certain location or purchasing patterns. That can produce unfair outcomes even when nobody intended them. Our guide to AI bias explains how this happens in more detail.

AI in claims

Claims is where customers feel insurance most directly, and where AI has some of its most visible uses.

Intake and triage

When a customer reports a loss, AI can capture details through an app or chatbot, classify the claim type and severity, and route it. Simple claims, like minor glass damage, can move toward fast-track handling. Complex or injury claims go to experienced adjusters.

Damage assessment

Computer vision models can review photos of vehicle or property damage and produce a preliminary estimate. Aerial imagery helps assess roofs and widespread catastrophe damage without sending an adjuster to every address.

Documentation and communication

Generative AI drafts claim summaries, letters and adjuster notes, and pulls key facts out of medical records, repair invoices and police reports. This saves time, but every output still needs review, since language models can state errors with confidence.

Where claims AI crosses a line

Using AI to pay claims faster is welcome. Using it to deny claims without meaningful human review is where trouble starts. Health insurers in particular have faced lawsuits and public criticism over allegations that algorithms were used to cut off or deny care. Many insurance laws already require fair, reasonable claim investigations, and regulators expect a human to stay accountable for adverse decisions.

Key takeaway: The safest pattern is “AI speeds up, people decide”. Let AI accelerate approvals and triage, but keep a qualified person responsible for denials, reductions and other adverse outcomes, and be able to explain why.

AI in fraud detection

Fraud is a large target. The Coalition Against Insurance Fraud’s 2022 report estimated that insurance fraud costs Americans about $308.6 billion a year across all lines, though the organization notes the true figure is impossible to measure precisely and some researchers dispute the methodology.

AI fraud tools typically do 3 things:

1. Anomaly detection. Flag claims that look unusual compared with similar ones, such as unusual timing, amounts or repair patterns.
2. Network analysis. Find links between claimants, providers, repair shops, phone numbers and addresses that suggest organized rings.
3. Text and image analysis. Spot reused photos, edited documents or suspicious language in claim narratives.

A fraud score should trigger investigation, not an automatic denial. False positives hurt honest customers, and a system that flags certain neighborhoods or groups more often can create discrimination problems of its own. Banks face similar trade-offs, covered in our guide on how banks use AI to catch fraud.

Insurers also face a newer problem: AI used against them. Generated photos, altered invoices and synthetic identities make document checks harder, which pushes fraud teams toward image forensics and verification steps.

AI in customer service and distribution

Chatbots and virtual assistants answer billing, coverage and claim-status questions, often around the clock. Agents and brokers use AI to summarize policies, compare quotes, draft client emails and prepare renewals. For a look at how independent agencies are adopting these tools, see how technology is changing insurance agencies.

The rules for customer-facing AI keep tightening. Several states now require disclosure when a consumer is talking to AI in certain contexts, and insurers remain responsible for what their chatbots say about coverage.

What regulators are focused on

Insurance in the US is regulated mostly by the states. Several frameworks matter most right now.

The NAIC AI model bulletin

The National Association of Insurance Commissioners (NAIC) adopted its Model Bulletin on the Use of Artificial Intelligence Systems by Insurers on December 4, 2023. It reminds insurers that decisions made or supported by AI must still comply with existing insurance laws, including those on unfair trade practices and unfair discrimination. It expects insurers to maintain a written AI systems program covering governance, risk management, internal controls and oversight of third-party vendors.

The bulletin is a model, not a law. Each state decides whether to issue it, sometimes with changes. By 2026, roughly half the states plus Washington, DC had adopted it or similar guidance. Check the NAIC’s adoption map and your state insurance department for the current list.

Colorado SB21-169

Colorado’s SB21-169, signed in July 2021, requires insurers to show that their use of external consumer data, algorithms and predictive models does not unfairly discriminate based on protected characteristics such as race. The Colorado Division of Insurance first applied its governance regulation (Regulation 10-1-1) to life insurers. An amended version, effective October 15, 2025, extended it to private passenger auto and health benefit plan insurers, with compliance reports due by July 1, 2026 and annually after that.

Broader state AI laws

General state AI laws also matter, and they are changing fast. Colorado, for example, replaced its 2024 AI Act with a narrower automated decision-making law that starts January 1, 2027. Our guide to US state AI laws tracks the details.

The EU AI Act

For insurers operating in Europe, the EU AI Act classifies AI used for risk assessment and pricing of individuals in life and health insurance as high-risk. After the 2026 Digital Omnibus amendments, high-risk obligations for these stand-alone systems apply from December 2, 2027. See our guide to the EU AI Act for US companies.

This is not legal advice. Rules vary by state and line of business and are still evolving, so confirm requirements with your state insurance department or counsel.

Where to start: a practical roadmap for insurers and agencies

Whether you are a carrier, an MGA or an agency, the steps are similar:

1. Inventory your AI. List every model, vendor tool and generative AI use, including AI built into software you already buy.
2. Rank by impact. Anything that affects eligibility, price, claim payment or fraud referral is high impact. Chat summaries and email drafting are lower.
3. Assign owners. Each system needs a named business owner and a technical owner.
4. Test for unfair outcomes. Check whether results differ across groups and whether inputs act as proxies for protected traits. Document the testing.
5. Keep people in the loop. Require human review for adverse decisions and make it easy to escalate.
6. Manage vendors. Ask vendors how models were built and tested, what data they use and how you can audit them. The NAIC bulletin makes clear that insurers stay responsible for third-party tools.
7. Write it down. A short written AI program, aligned with the NAIC bulletin, is the backbone. Our AI governance checklist offers a template.
8. Start with low-risk wins. Document extraction, call summaries and claim-status updates deliver value with less regulatory exposure than pricing models.

The bottom line

AI is now part of how insurance works, from the first quote to the final claim payment. Used well, it means faster quotes, quicker payouts and better fraud detection. Used carelessly, it can deny valid claims, discriminate through proxies and create regulatory problems.

The insurers that benefit most treat AI like any other material risk: inventoried, tested, documented and supervised by people who can explain the decisions. Start with the low-risk uses, build governance as you go and keep humans accountable for the decisions that matter most to customers.

Sources

• NAIC: Model Bulletin on the Use of Artificial Intelligence Systems by Insurers (adopted December 4, 2023) — AI systems program, governance and vendor expectations
• NAIC: Implementation of the AI model bulletin, state adoption map — which states have adopted the bulletin
• Colorado Division of Insurance: SB21-169, Protecting Consumers from Unfair Discrimination in Insurance Practices — law overview and rulemaking
• Colorado Division of Insurance: Notice of adoption, amended Regulation 10-1-1 — extension to auto and health insurers
• Coalition Against Insurance Fraud: The Impact of Insurance Fraud on the US Economy (2022) — $308.6 billion annual fraud estimate and its limits
• EUR-Lex: Digital Omnibus on AI, Regulation (EU) 2026/1744 — revised high-risk timelines under the EU AI Act