AI in Insurance Statistics 2026
By Axis Intelligence Research
Co-authors: Sarah Davis, Digital Finance and Fintech & Sarah Mitchell | Last updated: September 23, 2026 | License: CC BY 4.0
92% of US health insurers now use, plan to use, or are exploring AI and machine learning, according to the NAIC. Yet when Medicare Advantage insurers’ skilled nursing denials were appealed in June 2024, HHS OIG found 95% were overturned, and 82% of denials were never appealed at all. Axis Intelligence Research scores that distance as an Insurance AI Oversight Gap of 38.6.
Quick Answer: How Many Insurers Use AI in 2026?
According to NAIC line-by-line surveys, 88% of auto, 70% of homeowners, 58% of life and 92% of health insurers use, plan or explore AI/ML. In Europe, EIOPA reports 65% of insurers actively use generative AI. According to Axis Intelligence Research, only 30 of 51 US jurisdictions have an insurance-specific AI rule as of August 31, 2026.
Key Findings
- According to Axis Intelligence Research’s count of the NAIC implementation map, 30 of 51 US jurisdictions (58.8%) had an insurance-specific AI bulletin or regulation as of August 31, 2026, leaving 21 with none.
- Axis Intelligence Research finds that 21 of the 26 NAIC AI Model Bulletin adoptions happened in 2024; only one jurisdiction, Mississippi, adopted it in 2026.
- HHS OIG found Medicare Advantage insurers overturned 95% of appealed skilled nursing denials in June 2024, but Axis Intelligence Research calculates that only 17.1% of all denials were actually reversed, because 82% were never appealed.
- According to HHS OIG, naviHealth, a UnitedHealth Group contractor that processed half of all SNF admission requests, denied 14% of them, 1.27 times the 11% denial rate of insurers reviewing in-house, per Axis Intelligence Research.
- Axis Intelligence Research’s Insurance AI Oversight Gap (IAOG) reads 38.6 on a 0 to 100 scale in September 2026, measuring how far AI deployment across insurance lines runs ahead of state rules and consumer recourse.
How Many Insurance Companies Use AI? Adoption by Line of Business
The National Association of Insurance Commissioners is the only body that has surveyed US carriers on AI one line at a time, with regulators compelling responses. That makes its series the most defensible adoption baseline in the market, and it is the one this page builds on rather than vendor polls.
AI Adoption Rate by Insurance Line (NAIC Surveys)
| Line of business | Use, plan or explore AI/ML | Responding insurers | Report date | Source |
|---|---|---|---|---|
| Private passenger auto | 88% | 193 | Dec 2022 | NAIC |
| Homeowners | 70% | 194 | Aug 2023 | NAIC |
| Life | 58% | 161 | Dec 2023 | NAIC |
| Health | 92% | 93 | May 2025 | NAIC |
| Four-line mean | 77.0% | 641 | Sept 2026 | Axis Intelligence Research calculation |
The health figure needs a footnote the headlines skip. The NAIC health survey memo counts generalized linear models and generalized additive models as AI, which the auto and home surveys did not. So 92% is partly a definitional lift. The stricter “currently using” health figure is 84%, drawn from 93 companies across 16 states, per the NAIC Health AI/ML Survey Report.
The gap that matters sits between health and life. According to Axis Intelligence Research, health insurers report AI activity 34 percentage points above life insurers (92% versus 58%). Europe shows the same pattern: EIOPA found 50% of non-life insurers and 24% of life insurers were already using traditional AI in its 2024 digitalisation work. Life carriers lag on both sides of the Atlantic. We do not merge the US and EU figures, because the NAIC and EIOPA ask different questions of different samples.
Sarah Davis: Life insurance moves slowly for a reason you can find in the product itself. A whole life block priced a decade ago is still earning its spread today, and nobody rebuilds the engine on a policy with decades of duration left. Health plans reprice every year and adjudicate claims at enormous volume. The money is in the claims line, so that is where the models went first.
What Do Health Insurers Use AI For?
The NAIC health survey is the first regulator dataset that breaks AI down by specific function inside health plans. For individual major medical coverage, the top uses are the ones that decide whether care gets paid.
| Use case (individual major medical) | Insurers using or exploring AI | Source |
|---|---|---|
| Utilization management | 71% | NAIC Health AI/ML Survey, May 2025 |
| Prior authorization | 68% | NAIC Health AI/ML Survey, May 2025 |
| Disease management programs | 61% | NAIC Health AI/ML Survey, May 2025 |
| Provider fraud detection | 51% | NAIC Health AI/ML Survey, May 2025 |
| Claims fraud detection | 50% | NAIC Health AI/ML Survey, May 2025 |
| Sales and marketing | 45% | NAIC Health AI/ML Survey, May 2025 |
Utilization management and prior authorization both outrank fraud detection. That ordering reverses the industry’s usual talking point, which leads with fraud. According to Axis Intelligence Research, AI in US health insurance is first and foremost a coverage-decision technology.
Who Builds the Models? Third-Party AI in Insurance
Only 10% of health insurers’ AI models were built purely in-house, per the NAIC health survey memo. The rest involve outside vendors: 55% were developed internally with third-party components, 15% were built by a third party and 13% were built jointly with one. On the positive side, 85% of insurers said their vendor contracts contain no clause limiting disclosure to regulators. That matters for the vendor-oversight provisions discussed below, since an examiner cannot audit a model the contract hides.
How Often Are AI-Assisted Insurance Denials Overturned?
This is the question driving 2026 coverage of AI in health insurance, and the only hard answers come from federal oversight bodies with request-level data. Algorithms and human reviewers work side by side in these systems, so none of the figures below isolates an “AI-only” denial rate. What they show is how denial behavior changed as automation spread.
Medicare Advantage Post-Acute Denials Before and After Automation
The US Senate Permanent Subcommittee on Investigations reviewed more than 280,000 pages of internal documents from the three largest Medicare Advantage insurers. Its October 2024 staff report, Refusal of Recovery, found UnitedHealthcare’s prior authorization denial rate for post-acute care rose from 8.7% in 2019 to 10.9% in 2020 and 22.7% in 2022, while the company was working to automate reviews.
| Year | UnitedHealthcare post-acute denial rate | Source |
|---|---|---|
| 2019 | 8.7% | US Senate PSI |
| 2020 | 10.9% | US Senate PSI |
| 2022 | 22.7% | US Senate PSI |
| 2019 to 2022 multiple | 2.6x | Axis Intelligence Research calculation |
The Denial Funnel: What Happens to 1,000 Skilled Nursing Requests
In June 2026, HHS OIG published request-level data from 19 Medicare Advantage organizations for June 2024. They denied 12% of skilled nursing facility (SNF) admission requests, with insurer rates ranging from 0.4% to 23%. Enrollees appealed 18% of denials, and insurers overturned 95% of those appeals.
Axis Intelligence Research converted those rates into a funnel per 1,000 requests:
| Stage | Per 1,000 SNF requests | Basis |
|---|---|---|
| Denied | 120 | 12% denial rate (HHS OIG) |
| Appealed | 21.6 | 18% of denials (HHS OIG) |
| Reversed on appeal | 20.5 | 95% of appeals (HHS OIG) |
| Denied and never appealed | 98.4 | Axis Intelligence Research calculation |
The headline “95% overturned” describes a small, motivated subset. According to Axis Intelligence Research, just 17.1% of all SNF denials were reversed through appeal (0.18 x 0.95), and 98.4 of every 120 denials stood without challenge. We deliberately do not apply the 95% overturn rate to unappealed denials: families who appeal differ from those who do not, so that projection would be a guess dressed as a finding. It is logged as a declined calculation in our dataset.
Sarah Davis: In payments we call this a chargeback economy. If only one customer in five disputes a bad charge, a 95% win rate for disputers tells you the merchant knows exactly which charges will not survive scrutiny. The economics of a denial do not depend on being right. They depend on how few people push back.
Algorithmic Contractors Deny More Often
HHS OIG isolated one variable most studies cannot: who made the decision. naviHealth, a UnitedHealth Group subsidiary known for its nH Predict post-acute tool, processed half of all SNF admission requests in the sample and denied 14% of them. Insurers reviewing in-house denied 11%, and other contractors denied 9%. When enrollees appealed naviHealth denials, 97% were overturned.
| Reviewer | SNF denial rate | Multiple vs naviHealth | Source |
|---|---|---|---|
| naviHealth (contractor) | 14% | baseline | HHS OIG |
| MAO internal review | 11% | naviHealth is 1.27x | HHS OIG; Axis calculation |
| Other contractors | 9% | naviHealth is 1.56x | HHS OIG; Axis calculation |
The same OIG report found nursing home residents were denied SNF-level care 40% of the time, against 11% for all other enrollees. According to Axis Intelligence Research, that is a 3.6x denial multiple for the frailest enrollees in the program.
Long-Term Acute Care and Rehab Denials
A companion OIG report found the three largest Medicare Advantage insurers denied long-term care hospital and inpatient rehabilitation requests at higher rates than most peers. On appeal, insurers overturned 36% of LTCH denials and 43% of IRF denials, and IRF overturn rates ranged from 14% to 86% depending on the insurer. A 72-point spread for the same clinical category is not a medical disagreement. It is a process difference.
Is AI Coming to Traditional Medicare Prior Authorization?
Yes. The CMS WISeR Model (Wasteful and Inappropriate Service Reduction) uses AI and machine learning with human clinical review for selected services in traditional Medicare. It runs for six performance years, from January 1, 2026 to December 31, 2031, in six states: New Jersey, Ohio, Oklahoma, Texas, Arizona and Washington.
For the first time, algorithmic prior authorization is operating inside fee-for-service Medicare rather than only in private plans. The OIG funnel above is the natural benchmark for judging it: denial rates, appeal rates and overturn rates by reviewer. Axis Intelligence Research will track WISeR outcomes against that baseline as CMS releases data.
Which States Regulate AI in Insurance?
As of August 31, 2026, the NAIC implementation map lists 25 states and the District of Columbia as having adopted the NAIC Model Bulletin on the Use of Artificial Intelligence Systems by Insurers, first adopted by the NAIC on December 4, 2023. Four more states, California, Colorado, New York and Texas, run their own insurance-specific AI instruments. Of the four, only Colorado’s 3 CCR 702-10 is codified as a regulation.
Insurance AI Rules by Jurisdiction (Snapshot, August 31, 2026)
| Status | Jurisdictions | Count | Source |
|---|---|---|---|
| Adopted NAIC AI Model Bulletin | AK, AR, CT, DE, DC, HI, IL, IA, KY, MD, MA, MI, MS, NE, NV, NH, NJ, NC, OK, PA, RI, VT, VA, WA, WV, WI | 26 | NAIC |
| Own insurance-specific framework | CA, CO, NY, TX | 4 | NAIC |
| No insurance AI instrument listed | Remaining states | 21 | Axis Intelligence Research calculation |
The Adoption Stall
Axis Intelligence Research dated every adoption on the NAIC list. The pattern is stark: 21 jurisdictions adopted the bulletin in 2024, four in 2025 (Delaware, New Jersey, Wisconsin, Hawaii) and one in 2026 (Mississippi, July 22). According to Axis Intelligence Research, 80.8% of all adoptions came in the first full year, and the pace has since dropped to a trickle.
| Year | Bulletin adoptions | Source |
|---|---|---|
| 2024 | 21 | Axis Intelligence Research count of NAIC map |
| 2025 | 4 | Axis Intelligence Research count of NAIC map |
| 2026 (to Aug 31) | 1 | Axis Intelligence Research count of NAIC map |
Nearly every holdout state already has insurers deploying the same models used elsewhere. A carrier’s prior authorization engine does not change at the state line; only the documentation duty does. For the risk transfer side of this market, see our cyber insurance statistics and the CIPG protection gap.
How the EU AI Act Treats Insurance
Europe regulates by use case rather than by line. Under Annex III, point 5(c) of the EU AI Act, AI systems used for risk assessment and pricing of natural persons in life and health insurance are classified as high-risk, a scope EIOPA summarizes for the sector. Claims handling and property-casualty pricing sit outside that high-risk list. The American approach is the mirror image: US regulators have focused on claims and utilization decisions, while European law focuses on underwriting access.
How Fast Are European Insurers Adopting Generative AI?
EIOPA surveyed 347 insurance undertakings across 25 countries, per its press release. An April 2026 EIOPA speech put active generative AI use at 65% of undertakings, with most still at proof-of-concept stage.
Governance is moving faster than deployment. According to the EIOPA Generative AI Market Survey report, 49% of undertakings have a dedicated AI policy, up from 25% in 2023, and 69% of insurers already using generative AI have one.
| EU insurance AI metric | Value | As of | Source |
|---|---|---|---|
| Non-life insurers using AI (traditional) | 50% | 2024 | EIOPA |
| Life insurers using AI (traditional) | 24% | 2024 | EIOPA |
| Insurers actively using generative AI | 65% | 2025 survey | EIOPA |
| Insurers with a dedicated AI policy | 49% | 2025 survey | EIOPA |
| Insurers with a dedicated AI policy | 25% | 2023 | EIOPA |
| Generative AI users with an AI policy | 69% | 2025 survey | EIOPA |
For comparison with banks, which face similar high-risk provisions for credit scoring, see our AI in banking statistics. Finance and insurance also ranks among the leading US sectors in our AI adoption by industry data.
Will AI Replace Claims Adjusters and Underwriters?
The Bureau of Labor Statistics is unusually direct on this one. In its 2024 to 2034 projections overview, BLS names AI tools that assess photographs of property damage as a reason employment of claims adjusters, examiners and investigators is projected to fall 5.1%. The newer Occupational Outlook Handbook projection deepens that to a 6% decline from 2025 to 2035.
| Occupation metric | Value | Period | Source |
|---|---|---|---|
| Claims adjusters, examiners and investigators employed | 324,230 | May 2025 | BLS OEWS |
| Auto damage insurance appraisers employed | 11,560 | May 2025 | BLS OEWS |
| Claims occupation group, total employed | 335,790 | May 2025 | BLS OEWS |
| Projected employment change, claims | -5.1% | 2024 to 2034 | BLS |
| Projected employment change, claims | -6% | 2025 to 2035 | BLS |
| Projected annual openings, claims | 21,600 | 2025 to 2035 | BLS |
| Median annual wage, claims adjusters | $78,000 | May 2025 | BLS |
| Projected employment change, underwriters | -3% | 2024 to 2034 | BLS |
| Projected annual openings, underwriters | 8,200 | 2024 to 2034 | BLS |
According to Axis Intelligence Research, BLS deepened its projected decline for claims occupations by 0.9 points in a single projection cycle, one of the few cases where the agency explicitly attributes a revision to AI. Openings still run at 21,600 a year because retirements outpace the decline. The job shrinks; it does not vanish. Broader context is in our AI job displacement statistics.
Sarah Davis: Watch the auto damage appraiser line, not the headline occupation. At 11,560 workers it is small, and photo estimating hits it first. When a carrier lets the policyholder’s phone do the inspection, the cost that disappears is a salaried appraiser driving to a body shop.
The Insurance AI Oversight Gap (IAOG): Axis Intelligence Research’s Index
Adoption numbers alone tell a one-sided story. The Insurance AI Oversight Gap (IAOG) is an original Axis Intelligence Research index that measures how far AI deployment in US insurance runs ahead of the two checks on it: state supervision and consumer recourse. A reading of 0 means oversight fully matches deployment; higher readings mean a wider gap.
Formula:
IAOG = Deployment Breadth minus Oversight Score
Oversight Score = (0.5 x Supervisory Coverage) + (0.5 x Consumer Recourse)
| Component | Definition | Input value | Weight | Source |
|---|---|---|---|---|
| Deployment Breadth (D) | Mean share of insurers using, planning or exploring AI across auto, home, life and health | 77.0 | reference | NAIC surveys |
| Supervisory Coverage (S) | Share of 51 US jurisdictions with an insurance-specific AI instrument | 58.8 | 0.5 | NAIC map, Aug 31 2026 |
| Consumer Recourse (C) | Share of Medicare Advantage SNF denials that were appealed | 18.0 | 0.5 | HHS OIG, June 2024 data |
Calculation: Oversight Score = (0.5 x 58.8) + (0.5 x 18.0) = 38.4. IAOG = 77.0 minus 38.4 = 38.6.
According to Axis Intelligence Research, the September 2026 IAOG reading of 38.6 is the baseline reading of the index. Consumer recourse is the weakest component by far: supervisors now cover most jurisdictions on paper, but the people on the receiving end of automated decisions rarely contest them. Consumer Recourse uses the SNF appeal rate as a proxy because it is the only federally published, request-level appeal statistic tied to algorithmic review. The index moves when any input moves: a new NAIC line survey, a new state adoption, or new OIG or CMS appeal data. Each reading is dated and published on this page.
Sarah Davis: Regulators measure what insurers file. Nobody files the appeal that never happened. That silent 82% is the part of the ledger this index is built to keep visible.
Methodology
Axis Intelligence Research assembled this page exclusively from primary sources: the NAIC (line-of-business AI/ML surveys, health survey report and memo, model bulletin implementation map), HHS OIG (reports OEI-09-24-00330 and OEI-09-24-00331), the US Senate Permanent Subcommittee on Investigations, CMS, EIOPA and the Bureau of Labor Statistics. Every figure was retrieved on September 23, 2026 and logged with its URL in the downloadable dataset. Market-size forecasts from commercial research vendors are excluded by design, because their methods are not public.
Axis calculations shown on this page: the four-line NAIC mean (simple average of 88, 70, 58 and 92); the health-life gap (92 minus 58); jurisdiction counts from the dated NAIC adoption list (50 states plus DC as the denominator); adoptions by year from each listed adoption date; the SNF denial funnel (1,000 x 0.12, then x 0.18, then x 0.95); contractor denial multiples (14/11 and 14/9); the nursing home multiple (40/11); and the IAOG formula above. All arithmetic was re-run in Python before publication.
Scope notes: NAIC surveys were fielded in different years (2022 to 2025) and the health survey defines AI more broadly, so the four-line mean is a cross-line indicator, not a single-year census. PSI denial rates cover three insurers and do not isolate algorithmic decisions from human ones. OIG data cover one month (June 2024) and 19 insurers. The Senate PSI report PDF exceeds automated fetch limits; its figures were cross-checked against contemporaneous reporting of the report and are flagged for manual re-verification in each review cycle.
About This Dataset
The full dataset, including every figure above, the Axis calculations and one declined calculation with our reasoning, is available as ai-in-insurance-statistics-2026.csv under a CC BY 4.0 license. Each row carries its source organization, document, URL, retrieval date, a primary-source flag and a method note for calculated values. Updates are relevance-driven, never cosmetic: the page is revised when a new NAIC line survey, NAIC map change, OIG or CMS denial dataset, EIOPA survey or BLS projection is published.
Citation format: Axis Intelligence Research, AI in Insurance Statistics 2026, 2026.
Cite This Page
APA: Axis Intelligence Research, & Davis, S. (2026, September 23). AI in insurance statistics 2026: Adoption by line, algorithmic claim denials, regulation and the oversight gap. Axis Intelligence. https://axis-intelligence.com/ai-in-insurance-statistics/
MLA: Axis Intelligence Research, and Sarah Davis. “AI in Insurance Statistics 2026: Adoption by Line, Algorithmic Claim Denials, Regulation and the Oversight Gap.” Axis Intelligence, 23 Sept. 2026, axis-intelligence.com/ai-in-insurance-statistics/.
Chicago: Axis Intelligence Research and Sarah Davis. “AI in Insurance Statistics 2026: Adoption by Line, Algorithmic Claim Denials, Regulation and the Oversight Gap.” Axis Intelligence, September 23, 2026. https://axis-intelligence.com/ai-in-insurance-statistics/.
FAQ: AI in Insurance, Claims and Coverage Decisions
Can a health insurer deny a claim using AI alone?
No regulator publishes a count of AI-only denials. The NAIC Model Bulletin, adopted in 26 jurisdictions as of August 31, 2026, holds insurers accountable for decisions made with AI, including those from vendors. CMS’s WISeR Model in traditional Medicare pairs AI with human clinical review.
What percentage of Medicare Advantage denials are overturned on appeal?
It depends on the care setting. HHS OIG found insurers overturned 95% of appealed skilled nursing denials, 43% of inpatient rehab denials and 36% of long-term acute care denials in June 2024 data. Only 18% of skilled nursing denials were appealed.
Is it worth appealing an AI-assisted insurance denial?
For Medicare Advantage skilled nursing denials, federal data strongly favor appealing: 95% of appealed denials were overturned, and 97% of appealed naviHealth denials. Axis Intelligence Research calculates only 17.1% of all such denials were reversed, because most families never appealed.
Which line of insurance uses AI the most?
Health insurance, at 92% using, planning or exploring AI in the NAIC’s 2025 survey, followed by auto at 88%, homeowners at 70% and life at 58%. The health survey counted GLMs and GAMs as AI, which lifts its figure.
How many US states regulate insurers’ use of AI?
As of August 31, 2026, 25 states and DC had adopted the NAIC AI Model Bulletin, and California, Colorado, New York and Texas had their own instruments. Axis Intelligence Research counts 30 of 51 jurisdictions covered and 21 without an insurance-specific AI rule.
Does the EU AI Act classify insurance AI as high-risk?
Partly. Annex III, point 5(c) classifies AI used for risk assessment and pricing of individuals in life and health insurance as high-risk. Claims handling and property-casualty pricing are not on that list.
How many claims adjuster jobs will AI eliminate?
BLS projects employment of claims adjusters, appraisers, examiners and investigators will fall 6% from 2025 to 2035, citing AI photo-based damage assessment in its prior cycle. About 21,600 openings a year remain, mostly from retirements.
What is the Insurance AI Oversight Gap (IAOG)?
The IAOG is an Axis Intelligence Research index comparing AI deployment breadth across US insurance lines with state supervisory coverage and consumer appeal rates. Its September 2026 baseline reading is 38.6 on a 0 to 100 scale, where higher means oversight trails deployment further.