Insurance Fraud Statistics 2026: Claims, Costs and AI Scams

In June 2026, Allianz described a claim involving a van with a cracked bumper and a repair invoice for about £1,000. Investigators traced the photograph to the owner’s own business social media page. The damage had been added using AI, and the accident had never happened.

For claims teams, that creates an uncomfortable question: what does a convincing photograph actually prove?

The latest insurance fraud statistics show the scale of the wider problem, while recent insurer reports reveal how it is changing. Exaggerated losses, false applications and organised claims networks remain central concerns. AI gives fraudsters additional ways to make their stories look credible.

Below, we examine the figures, the emerging tactics and what insurers can do with this information.

Research updated 8 September 2026. Reporting periods are stated alongside the figures.

Insurance fraud statistics at a glance

These figures measure different things: estimated economic costs, detected or suspected claims, database filings and survey responses. They should not be added together or treated as directly comparable fraud rates.

How much does insurance fraud cost?

The Coalition Against Insurance Fraud’s $308.6 billion annual US estimate remains widely cited, including by the NAIC in its June 2026 overview. The underlying research dates to 2022. It is not a newly measured total for 2026.

Its scope is broad, covering property and casualty, workers’ compensation, premium avoidance, life insurance, healthcare, public medical programmes and auto theft. That breadth helps explain why it is much larger than figures focused on a single insurance line.

For a more recent view of enforcement activity, the US Department of Justice’s June 2026 healthcare fraud takedown resulted in charges against 455 defendants, including 90 doctors and other licensed medical professionals. The alleged schemes involved more than $6.5 billion in false claims.

Those are allegations attached to a coordinated enforcement action. They do not represent a national annual loss total, or a finding that every billed dollar was paid.

For insurance leaders, the useful distinction is between the amount attempted, the amount prevented and the amount ultimately lost.

UK claims fraud: motor leads, but exaggeration cuts across products

The ABI’s November 2025 release reported £1.16 billion in fraudulent general insurance claims detected during 2024, up 2% year on year. The number of detected claims increased faster, rising 12% to more than 98,400.

Its breakdown shows where the activity sat:

• Motor insurance: 51,700 detected fraudulent claims worth £576 million, about 53% of detected cases.
• Property insurance: 18,700 detected fraudulent claims worth £189 million, with case numbers up 11%.
• Exaggerated loss: the most common type, accounting for £466 million across the relevant claims.

“With insurers continuing to detect over £1bn worth of bogus claims, the fight must continue.”
Mark Allen, head of fraud and financial crime, Association of British Insurers

A genuine incident can still contain a fraudulent component. The collision or the escape of water may have happened, while the claimed damage, replacement values or associated expenses have been inflated.

That makes investigation more demanding than establishing whether an event occurred.

What insurers’ 2025 results reveal

Recent company disclosures give a more current view of insurance fraud trends, although each reflects its own portfolio and detection practices.

Admiral detected £86.8 million in fraudulent motor, home and travel claims in 2025, against £50.9 million in 2024. The company linked the increase to both claim exaggeration and improved detection.

That second factor deserves attention. A higher detected-fraud total can mean controls are finding activity that was previously missed.

Aviva reported more than 18,400 suspect claims worth £233 million for 2025, including Direct Line brands for the first time. Comparisons with earlier Aviva totals therefore need to account for that expanded reporting scope.

For benchmarking, ask what changed beneath the headline: portfolio size, business mix, investigation capacity, reporting definitions, or the underlying fraud itself.

Usually more than one of those needs examining.

AI-generated damage images are already appearing in claims

The fabricated van damage described earlier is one example of a broader evidence problem.

Allianz has also reported AI-assisted identity and document manipulation.

“This is a wide-scale problem, as we’re seeing it across all our products in both commercial and personal lines.”
Matt Crabtree, Head of Financial Crime Intelligence and Investigation Strategy, Allianz UK

Admiral has described a designer watch made to look damaged with an AI filter. Its other examples include conventional photo editing and images of damaged luggage taken from the internet. Those illustrate several kinds of manipulated evidence, and they should not all be labelled generative AI.

For claims operations, three scenarios deserve separate handling:

• Invented damage: an image depicts a loss that never occurred.
• Exaggerated damage: a real incident is supported by imagery that overstates its severity.
• Misrepresented evidence: a genuine photograph or document is presented as proof of a different incident, item or expense.

The distinction changes the investigation. An authentic photograph can still be irrelevant to the claim, and a suspicious image may need corroboration from other evidence before any conclusion is drawn.

Insurance Europe’s March 2026 submission on online fraud identifies fabricated images, reports and documents as an emerging challenge for insurers.

What AI insurance fraud actually means

AI insurance fraud is the use of AI-generated or AI-altered material to mislead an insurer during an application or a claim, as Insurance Europe frames it. That material might be a photograph, an invoice, an identity document or a written account of an incident.

The underlying deception is often familiar. What changes is how the supporting evidence gets produced.

Generative AI can create new content or modify existing material, which gives fraudsters another way to make a false account look credible. Beyond damage imagery, Aviva reports claims involving AI-generated accident scenes and manipulated documentation. Allianz has uncovered an application carrying an AI-created image of a driving licence, attached to a stolen identity.

For claims teams the question is wider than whether an image looks realistic. Investigators also have to establish whether it depicts the insured item, whether it relates to the reported incident, and whether it agrees with the rest of the evidence.

How widespread is AI insurance fraud?

Verisk’s March 2026 study surveyed 1,000 US adult consumers who owned a smartphone, computer or tablet, and 300 US insurance claims professionals at manager level or above. Fieldwork ran from December 2025 to January 2026.

Among the claims professionals, 99% reported encountering manipulated or AI-altered documentation, 76% said submissions had grown more sophisticated over the preceding year, and only 32% were very confident about identifying deepfakes.

Among consumers, 36% said they would consider a small edit to strengthen a claim even if it broke insurer rules. 13% considered creating a photograph of damage that never occurred acceptable.

These insurance fraud statistics describe reported experiences, confidence and attitudes. They do not establish the percentage of claims containing AI-generated evidence, or how many consumers have actually submitted it.

Encountering manipulated documentation is different from measuring how common it is across a portfolio.

The evidence supports taking the threat seriously while keeping those distinctions clear.

Fraud prevention also starts before a claim exists

Claims imagery attracts the attention, but fraudulent applications and identity misuse create exposure earlier in the lifecycle.

Cifas reported that insurance identity fraud filings rose 26% in 2025, passing 16,000 cases. Those are cases recorded by its members rather than every incident in the UK market.

Aviva’s UK general insurance business, excluding Direct Line, also reported stopping more than 105,000 fraudulent insurance applications in 2025. Identified ghost-brokered policies rose 7% year on year.

For insurers distributing through brokers, aggregators and embedded partners, that argues for reviewing application controls alongside claims controls. Rules you can change when a pattern shifts are worth more here than an annual model refresh, which is an argument for holding them in an automated underwriting system a business user can edit.

Does the identity information stay consistent? Are changes to customer details properly recorded? Can the claims team see how the policy was originated and amended?

A claim should arrive with its policy history available for review.

AI can help investigators connect the evidence

The same advances creating opportunities for fraud also give insurers new ways to investigate it. AI can support claims analysis, surface suspicious patterns and help teams prioritise what to look at next. Aviva says it combines advanced analytics and AI-enabled tools with human oversight to identify suspicious claims earlier.

For an insurer the value goes beyond spotting a manipulated photograph. A claim can look entirely plausible on its own, while its connections to previous claims, other policyholders or particular businesses are the thing worth investigating.

Bringing those connections into the handler’s view is what makes the available evidence useful.

In October 2025 Zurich announced an expanded approach using Quantexa’s platform to connect claims information with sources including the Claims and Underwriting Exchange, the Insurance Fraud Register and Companies House. The system highlights relationships between people, addresses, businesses and claims, giving handlers material to support a referral.

“Importantly, the real-time nature of the platform means we can reinforce our defences against fraudsters without holding up processes for genuine claimants.”
Scott Clayton, Head of Claims Fraud, Zurich

Cross-insurer connections can be especially valuable. In a case presented through Insurance Europe’s 2026 Insurance Crime Platform, investigators identified more than 67 linked motor cases across three countries and nine insurers. Repeated participants and accident patterns exposed the coordinated activity.

An image check examines one piece of evidence. Linking claims and identities can reveal relationships that stay invisible inside an individual file.

Both belong in the process.

For insurers building AI-supported detection, the practical aim is better evidence and clearer priorities for investigators. A suspicious finding still needs assessing, and a genuine claim still needs a straightforward route to settlement.

Which insurance fraud statistics should claims leaders track?

Industry figures set the context. Operational measures show whether your own response is working.

Agree consistent definitions while you are at it. A suspicious submission, an investigation referral and a substantiated finding are three different stages, and conflating them makes the trend meaningless.

For AI-assisted checks, keep the submitted evidence and record why the case was flagged, what corroboration was obtained and how the decision was reached. An insurance audit trail generated as a by-product of running is what makes a declination defensible later, and it is the same discipline the wider insurance claims data problem needs.

A fraud score should direct attention. The decision still needs an evidential basis.

Building fraud controls into everyday insurance operations

The practical challenge is assessing increasingly convincing evidence without making the claims experience worse for everybody else.

That needs claims handlers, fraud specialists and underwriters looking at the same history, and a clear route for referring a case, requesting more evidence, recording findings and putting a genuine claim back into normal processing.

On the platform question the honest position is narrow. Openkoda is not a fraud analytics product and would lose a feature comparison with one. What it offers is that referral rules, application questions and the claim record are configuration you own rather than a change request against someone else’s release calendar, which matters because fraud patterns move faster than a development calendar does. Our guide to AI-driven insurance analytics covers how the data side supports claims, underwriting and fraud work.

The next improvement may be a better detection tool. It may equally be making sure the evidence you already hold reaches the person who has to assess it.