54 AI Insurance Fraud Statistics for 2026: Edited Photos, Deepfakes and the Detection Gap

More than one in three consumers say they would consider editing a claim photo to strengthen their case. Almost every insurer surveyed in 2026 has already seen manipulated or AI-altered documentation. And fewer than a third of those insurers feel very confident they could spot a deepfake.

This page collects 54 statistics on AI and photo manipulation in insurance fraud, drawn from Verisk, Aviva, Allianz, the ABI, the Coalition Against Insurance Fraud, Sumsub, Deloitte, CCC and SAS. Every figure links to the page where we found it. We wrote it for the claims, counter-fraud and SIU leaders who have to decide what a claimant's photo is worth, and for anyone writing about the topic who needs a number with a source behind it.

Our team builds Venta Capture, a product of VentaVid, which records claim evidence in the session on the claimant's own phone instead of accepting whatever file they upload. If the numbers below describe your referral queue, start a free account and send yourself a capture link.

Key highlights

• 36% of US consumers would consider digitally altering a claim image or document, rising to 55% of Gen Z (Verisk, March 2026)
• 99% of insurers have encountered manipulated or AI-altered documentation, and 98% say AI editing tools are driving a rise in digital fraud (Verisk)
• 66% of insurers believe digital media fraud goes undetected often or very often (Verisk)
• Only 32% of insurers are very confident they could identify a deepfake (Verisk)
• Aviva detected a record £233 million of suspect claims in 2025, with the value of motor fraud up 39% (Aviva, June 2026)
• Deepfake attempts in the UK rose 94% in 2025 (Sumsub)
• Insurance fraud costs the US an estimated $308.6 billion a year (Coalition Against Insurance Fraud)

What insurance fraud costs before AI enters the picture

The AI numbers only make sense against the base rate. These are the totals the industry already carries.

Insurance fraud costs the United States an estimated $308.6 billion every year. The figure comes from the Coalition Against Insurance Fraud's 2022 study with Colorado State University Global, which replaced the 1995 estimate of $80 billion and covers every line of insurance. (Coalition Against Insurance Fraud)

Property and casualty fraud alone accounts for $45 billion of that total. The same study puts fraud at about 10% of property-casualty losses, the multiplier the Insurance Information Institute has used for years. (Coalition Against Insurance Fraud, 2022 report)

UK insurers detected £1.16 billion of fraudulent general insurance claims in 2024, up 2% on the year before. The count rose faster than the value: 98,400 fraud-related claims, a 12% increase on 2023. (ABI figures via Insurance Journal)

Exaggerated loss is the most common type, worth £466 million in 2024, up 10%. This matters for the photo question: an exaggerated claim is a genuine loss with the damage inflated, which is exactly what an edited photo is good for. (ABI via Insurance Journal)

Aviva uncovered more than 18,400 suspect claims worth £233 million in 2025, or £638,000 of fraud every day. It is the first year Aviva has reported combined figures including Direct Line, so the record comes with that footnote. (Aviva, June 2026)

Allianz UK prevented nearly £174 million of fraud across 34,200 cases in 2025, a 10.5% rise on the £157 million of 2024. That works out at £3.3 million recouped every week. (Insurance Age)

22% of drivers admit to lying to their insurer, and 78% of consumers say they are concerned about insurance fraud. Both figures sit on the Coalition's fraud statistics page. (Coalition Against Insurance Fraud)

Two things stand out. Detected fraud keeps hitting records, and the growth is in the count of smaller, opportunistic cases rather than in organised rings. That is the population AI editing tools serve.

How willing consumers are to edit a claim photo

Verisk surveyed 1,000 US consumers and 300 claims professionals between December 2025 and January 2026. It is the first large study to ask policyholders directly whether they would touch a claim image.

36% of consumers would consider digitally altering a claim image or document, even knowing it breaks their insurer's rules. (Verisk, March 2026)

55% of Gen Z and 49% of millennials would consider it. Only 28% of Gen X and 12% of baby boomers say the same. The generational slope is steep enough that the average claimant of 2035 looks very different from the one of 2020. (Verisk)

41% of consumers know someone who has used AI editing tools to alter or create a photo, video or document for financial gain, including insurance claims, product returns and online sales. Among Gen Z the figure is 64%. (Verisk)

44% of consumers who have used AI editing tools describe their results as "very realistic". (Verisk)

62% of consumers believe people use AI tools to manipulate claim documents often or very often. (Verisk)

52% say adjusting brightness or contrast to make damage easier to see is acceptable, and 49% are comfortable cropping out background elements. (Verisk)

41% consider flipping, rotating or repairing a blurry photo acceptable. (Claims Journal)

15% think exaggerating damage is acceptable, and 13% say creating an image of damage that never occurred is acceptable. (Verisk)

53% of insurers believe at least half of the policyholders who alter claim media do not realise the edit may count as fraud. (Claims Journal)

The gradient from "brightness" to "damage that never occurred" is the whole story. Most claimants sit in the grey band, editing for clarity, and the same tool that lifts contrast will also paint a dent. A claims process that accepts uploaded files cannot tell the two apart.

What insurers say they are seeing

99% of insurers have encountered manipulated or AI-altered documentation. (Verisk)

98% agree that AI-powered editing tools are driving a rise in digital media fraud. (Verisk)

76% say AI-altered claim submissions have become more sophisticated in the past year. (Verisk)

Allianz UK saw a 300% increase in cases where apps were used to distort real-life images, videos and documents from 2022 into 2023. This is the figure most often quoted for shallowfakes: real media edited with ordinary software, no AI required. (Allianz UK, April 2024)

Aviva reports a growing number of claims supported by AI-generated images and manipulated documents, particularly in motor. The insurer describes fraudsters fabricating accident scenes and damage imagery to support false or exaggerated claims. Aviva has not published a count, so treat this as a direction rather than a rate. (Aviva, June 2026)

Fraud inside Aviva's home insurance book rose 15% in 2025, often hidden within otherwise legitimate claims where customers exaggerate damage, repairs or contents values. (Aviva)

One case Allianz published shows the pattern at its smallest: a tradesman's van, photographed for his business page, copied from social media, given a cracked bumper in an editing app and filed with a repair invoice of around £1,000. Nothing in the photo was fake except the damage. We walk through more of these in shallowfake insurance claims and AI-generated insurance fraud.

Motor fraud in the UK: where edited photos land first

Motor is the line where a photo most often decides the payout, so it is where image manipulation shows up first.

UK insurers detected 51,700 motor insurance scams worth £576 million in 2024, 5% more than in 2023 and 53% of all detected fraudulent claims. (ABI via Insurance Journal)

Motor accounts for more than seven in ten of the fraudulent claims Aviva detects. (Aviva)

The value of motor fraud Aviva detected rose 39% in 2025, as fraudsters moved away from staged collisions and towards exaggerated claims for vehicle damage, repair costs, credit hire and injury. (Aviva)

Liability fraud value at Aviva rose 32% on stable case volumes, driven by exaggerated loss of earnings, rehabilitation and injury claims. (Aviva)

52% of UK drivers aged 18 to 34 have posted a picture of their car online, and 73% had no idea it could expose them to insurance fraud. Allianz's point is that a public photo of your car is raw material for someone else's edited claim. 66% said they would be more cautious in future. (Allianz UK)

Motor fraud adds an estimated £50 to £60 to every UK policy. (Insurance Business UK)

Aviva stopped more than 105,000 fraudulent insurance applications in 2025, with ghost broking up 7% year on year. (Aviva)

UK insurers as a whole prevented 684,800 fraudulent applications in 2024, up 7.4%. (ABI via Insurance Journal)

The shift from staged collisions to inflated repair claims is the important line in Aviva's release. A staged crash needs people and cars. An inflated repair claim needs a photo and ten minutes.

Deepfakes and AI-generated documents in the wider fraud picture

Insurance is one target among many. These figures come from identity verification and financial services, and they show how fast the tooling is moving.

Sophisticated fraud rose 180% globally in 2025, defined by Sumsub as multi-step, coordinated attacks combining several advanced techniques in a single verification attempt. The report draws on more than 4,000,000 fraud attempts analysed between 2024 and 2025. (Sumsub, Identity Fraud Report 2025-2026)

Deepfake attempts in the UK rose 94% in 2025. France saw 96%, Spain 84% and Germany 53%. (Sumsub via PR Newswire)

72% of EU companies expect more sophisticated attacks using AI, particularly deepfakes and AI-generated identity documents. (Sumsub via PR Newswire)

One in five fraudulent verification attempts involved edited or forged ID documents, and 2% of falsified documents globally were created with generative AI tools, a category Sumsub expects to grow by double digits in 2026. (Sumsub via PR Newswire)

Deloitte expects generative AI to push US fraud losses from $12.3 billion in 2023 to $40 billion by 2027, a 32% compound annual growth rate. The forecast covers financial services broadly, not insurance alone. (Deloitte Center for Financial Services)

Deepfake incidents in fintech rose 700% in 2023. (Deloitte)

The UK government predicted eight million deepfakes would be shared in 2025, up from 500,000 in 2023. (Insurance Business UK)

Keep the proportions in mind. Fully synthetic media is still a small share of document fraud. Edited real media is the volume problem, and it is the one a claims team meets every day.

The detection gap

This is the section to read twice. Insurers are confident about the easy cases and much less so about the ones that are growing.

66% of insurers believe digital media fraud goes undetected often or very often across the industry. (Verisk)

58% are very confident they can detect edits made to real photos or videos. (Verisk)

43% feel very confident assessing the authenticity of digital media at scale. (Verisk)

Just 32% are very confident they could identify a deepfake. (Verisk)

39% point to insufficient integration between fraud tools and claims systems, 38% say their detection tools miss too many altered images, 35% report false positives and 34% struggle to keep up with evolving techniques. (Claims Journal)

Only 7% of anti-fraud professionals say their organisation is more than moderately prepared to detect or prevent AI-driven fraud. Among insurance respondents specifically, none expressed more than moderate confidence. (SAS and ACFE survey, via SAS, May 2026)

80% of insurers use predictive modelling to detect fraud, up from 55% in 2018. The tooling has grown; the confidence has not kept pace. (Insurance Information Institute)

Read those together and the picture is this: detection after the fact is losing ground to editing before the fact. The Verisk respondents are not complaining about a lack of tools. They are describing tools that see a file after the edit has already been made, with no way to know how the file came to be. That is a provenance problem, not a pattern-matching problem.

If your team is weighing where to put next year's counter-fraud budget, book a 20-minute demo and we will show what a claim file looks like when the evidence is recorded in the session and sealed on receipt.

Photo-based claims handling as the attack surface

The reason edited photos pay is that photos now settle a large share of claims without anyone seeing the vehicle.

Photo estimating accounted for 26.4% of US auto claim inspections in 2025, up nearly a percentage point, with direct repair programmes at 46.7%. (CCC Crash Course 2026, via Claims Journal)

Total loss frequency reached 23.1% of claims, a record. (CCC via Claims Journal)

The average repair cost in 2025 sat between $4,500 and $5,000, against roughly $2,500 in 2010. Higher severity means a small percentage inflation on a photo estimate is worth more than it used to be. (CCC via Claims Journal)

45% of insurers expect tighter documentation or proof-of-loss requirements for claimants within three to five years, and 35% expect longer claim cycle times. The industry sees the trade-off coming: more scrutiny at intake, slower files. (Verisk)

36% of consumers worry that legitimate claims will be delayed or denied if altered documents are mistakenly flagged, and 69% expect fraudulent claims to raise premiums for everyone. (Verisk)

A quarter of inspections on a photo, severity doubling in 15 years, and a claimant base where one in three would consider an edit. Those three numbers, side by side, explain why photo evidence in insurance claims has become a board-level topic rather than an SIU footnote.

What insurers are doing about it

54% of insurers are increasing internal training, 51% have issued new guidance for adjusters, 48% are running internal audits and 47% are investing in new fraud detection technology. (Claims Journal)

65% use automated AI-based detection tools from third parties, 50% use internally developed AI tools and 44% still rely on manual review. (Verisk and Claims Journal)

48% expect increased adoption of technology solutions to offset digital media fraud over the next three to five years, and 36% expect greater operational strain on claims teams. (Verisk)

Allianz UK's fraud savings rose 29% in 2023, the year it first flagged app-manipulated media as "the latest big scam to hit the insurance industry". (Allianz UK)

Aviva secured more than 37 years of custodial and suspended sentences for the most serious fraud offences in 2025. (Aviva)

Training, guidance, audits and detection software all act on the file after it arrives. None of them change what the file is.

What this means for claims, fraud and SIU leaders

Three conclusions follow from the data, and none of them require you to believe a vendor.

The editing window is the problem, not the editor. Every shallowfake and most AI-assisted edits happen in the gap between the moment a photo is taken and the moment it is uploaded. Verisk's 36% and Allianz's 300% both sit in that gap. Close it, and the majority of opportunistic manipulation has nowhere to happen. Detection tools that inspect a finished file cannot close it; only the intake design can.

Provenance beats forensics for the volume cases. Pixel forensics and reverse image search still matter for organised fraud. For the 15% who think exaggerating damage is fine, what you need is to know when, where and on which device the image was made, with a server-side receipt time you did not have to take on trust. That is what evidence integrity means in practice, and it is cheaper than a second detection layer.

Honest claimants pay for weak intake too. 36% of consumers already fear a legitimate claim being flagged by mistake, and 35% of insurers expect cycle times to lengthen. A capture process that produces a clean, sealed file on the first attempt removes the photo chase-up for the honest majority and gives the SIU a smaller, better-documented referral set. That is how claims cycle time comes down while scrutiny goes up.

This is the design behind Venta Capture for remote claim inspection. The claimant gets a link, no app and no account, and records the damage in the session on their own phone with step-by-step guidance. The submission arrives digitally sealed, with timestamps, device context and a hash your team can verify externally, and any extra upload the claimant adds is labelled as unverified with its signals alongside it. The product tells your handler what it can and cannot prove. The decision stays with the handler.

Start a free account and run one motor damage flow past your own SIU, or book a demo if you would rather see it on your own claim types first.

Frequently asked questions

How many consumers would edit an insurance claim photo?

In Verisk's 2026 State of Insurance Fraud study, 36% of US consumers said they would consider digitally altering a claim image or document to strengthen their case, even knowing it breaks their insurer's rules. Among Gen Z the figure is 55%, among millennials 49%, and it falls to 12% for baby boomers.

What percentage of insurers have seen AI-altered claim documents?

99% of the 300 claims professionals Verisk surveyed said they had encountered manipulated or AI-altered documentation, and 98% said AI editing tools are driving a rise in digital media fraud. 76% said altered submissions had become more sophisticated in the past year.

How much does insurance fraud cost?

The Coalition Against Insurance Fraud estimates $308.6 billion a year in the United States across all lines, with $45 billion in property and casualty. In the UK, the ABI reported £1.16 billion of detected fraudulent general insurance claims in 2024, and Aviva alone detected £233 million in 2025.

What is the difference between a shallowfake and a deepfake in insurance?

A shallowfake is a real photo, video or document edited with ordinary software: a dent painted onto a genuine car photo, a date changed on an invoice. A deepfake is synthetic media generated by AI. Allianz UK reported a 300% rise in app-manipulated media between 2022 and 2023, and shallowfakes remain the larger volume because they need no special skill.

How confident are insurers in detecting deepfakes?

Not very. Verisk found that 58% of insurers are very confident they can detect edits to real photos, but only 43% are very confident assessing authenticity at scale and just 32% are very confident they could identify a deepfake. 66% believe digital media fraud goes undetected often or very often.

Is the "20 to 30% of claims contain AI-altered media" figure reliable?

We could not find a primary source for it. It circulates on vendor blogs without a named study, sample or date, so we left it out of this page. The closest verified figures are Verisk's 36% consumer willingness and 99% insurer exposure rates, which measure different things.

Does guided capture stop AI insurance fraud?

No single control stops fraud, and a recording made in the session can still show a staged scene. What guided, sealed capture removes is the editing window between recording and submission, which is where shallowfakes and most AI-assisted edits are made. It also gives the handler a verified receipt time, device context and an integrity hash, so the file can stand up later in a dispute.

Methodology and update notes

We checked every source page directly in September 2026 and only included figures we saw on the page linked. Where a primary page blocked automated access (the ABI's November 2025 release and CCC's Crash Course 2026 page), we cite the trade-press article that reproduces the figures and name the original publisher. We excluded figures from vendor blogs with no primary source, including the widely repeated claim that 20 to 30% of claims contain AI-altered media. Statistics from identity verification and financial services (Sumsub, Deloitte) are labelled as such and should not be read as insurance claim rates. If a source updates a figure, we will update this page; the date in the frontmatter is the last review.

For the operational side of the same problem, see insurance claim documentation that survives a dispute and FNOL automation.