Image fraud in insurance. Global trends 2026.
The claim photo is the evidence that carries the most weight. Today it can be edited in seconds from a phone. In this report we review the global insurance fraud figures and point out what works to detect it.
Vouch teamOctober 21, 202610-minute read
The numbers
of U.S. consumers would consider altering an image or document to strengthen a claim. Among Gen Z, 55%.1
Survey of 1,000 consumers and 300 claims professionals, Dec 2025–Jan 2026.of claims professionals feel very confident they can spot a deepfake. 98% believe AI editing is driving fraud.1
Survey of 1,000 consumers and 300 claims professionals, Dec 2025–Jan 2026.in fraudulent claims detected by UK insurers in 2025. More than half of the cases were motor.2
Association of British Insurers data: 93,900 cases.in proven fraud in Brazil in 2024, out of R$ 5.4 billion in suspicious claims.3
CNseg Fraud Quantification System.For scale: in the United States, the Coalition Against Insurance Fraud estimates insurance fraud costs USD 308.6 billion a year.4
1. What changed: the photo no longer proves the damage
For years, the claim photo was the most reliable evidence. With digital self-inspection it often became the only one: the policyholder takes the photo, uploads it and the claim moves forward without anyone seeing the car or the house.
Today that photo can be retouched in seconds from a phone. The industry calls these images shallowfakes: real photos with a dent, a crack or a stain added. Documented cases show it is no longer an isolated problem.
- May 2024United KingdomAllianz UK reports that cases involving images, videos and documents altered with apps rose 300% between 2021-22 and 2022-23, and warns about shallowfakes.5
- Apr 2025BrazilFenSeg warns of fraud using AI-generated images and even toy model cars photographed to look like real vehicles.6
- 2025ArgentinaIn its annual anti-fraud contest, CESVI flags AI image tampering as a rising trend, along with edited repair quotes and fake medical certificates.7
- Feb 2026ArgentinaAdjusters and lawyers are already receiving AI-created claim images that pass basic checks.8
- Mar 2026United StatesVerisk publishes its study on AI editing in claims: consumers and insurers agree the problem is growing.1
- Apr 2026United KingdomAdmiral detects £86.8 million in fraud in 2025, up 71% on 2024. Cases include photos of damaged cars edited with a different number plate to claim twice.9
- Jun 2026United KingdomAviva uncovers 18,400 suspect claims worth £233 million in 2025 and sees more claims backed by AI-generated images and manipulated documents, especially in motor.10
- Jun 2026United KingdomAllianz UK warns that car photos posted online with the number plate visible are being used to claim for damage created with AI.11
- Jul 2026BrazilProven fraud rises from 1.8% to 3.3% of earned premiums between 2021 and 2025, and digital alteration of accident photos emerges as a recurring pattern.12
2. How the fraud works
An altered photo shows up in almost every tactic, from the invented claim to the exaggerated one.
Added damage
A sound car or item shows a dent, crack or water stain that does not exist. Allianz found a van with a crack added with AI, submitted with a repair invoice.11
Exaggerated damage
The loss is real, but the photo makes it look worse. In Spain, AXA reported that 60% of its 2024 fraud cases were policyholders with a real claim who tried to add damage that did not belong to it.13
Recycled photo
The image is taken from social media, sales or salvage sites, or reused in another claim. Car photos posted with the number plate visible are already being used this way.11
Swapped number plate
The same photo of a damaged car is edited with a different plate to claim twice. Admiral caught and denied cases like this before paying.9
Fabricated item
In claims for damaged or stolen items, the evidence can be an image altered with AI, like the designer watch made to look “damaged” with a filter that Admiral detected.9
Staged scene
The accident never happened. In Brazil, fraud using toy model cars photographed to look like real vehicles has been reported.6
3. Who pays
Fraud is not paid by the insurer alone. It is passed on to every policyholder’s premium and slows down honest claims, because every extra check applies to everyone.
And the problem is growing in value more than in volume. In the UK, detected cases fell in 2025, but the average fraudulent claim reached £14,300, the second highest on record.14 Every photo approved without analysis weighs more.
A quick estimate
Fraud loss = claims with photos × share with altered images × average payout. Replace the values with your own book.
Estimated lossUSD per month · USD per year
Example values, not market data. Does not include adjuster costs or handling time.
4. Why current controls fall short
The adjuster’s eye
Only 32% of claims professionals feel very confident they can spot a deepfake.1 AI-generated images don’t leave the editing traces that traditional tools used to leave.8
Self-inspection without analysis
The photo the policyholder uploads replaces the in-person inspection. If nobody analyzes it, approving the photo means approving the claim.
Metadata as the only proof
File data helps, but it is lost when the photo is forwarded by chat or screenshotted, and an AI-generated image may not have it at all.
Controls that look elsewhere
Traditional fraud models cross-check histories, amounts and claim patterns. They are useful, but they don’t analyze whether the image was edited, which is exactly what is being faked today.
More sophisticated fakes
76% of insurers say claims with altered images became more sophisticated over the past year.1
5. What works: a layered approach
No single signal is enough. What works is analyzing every photo with several layers at once and saving adjusters for the doubtful cases.
Image analysis
Look for signs of generation or editing in the damage and the number plate: edges, reflections, lighting and textures that don’t match a real photo. The file’s metadata helps, but it can’t be the only check.
Image history
Know whether that photo has appeared before: on social media, on sales or salvage sites, or in another claim, even if cropped or rotated.
Claim context
Cross-check the photo with the policyholder’s and the policy’s history, claim frequency and what is being declared.
Human review
A specialist reviews only the gray-area cases, with the evidence already organized, and records the reasoning for audits, disputes and the regulator.
6. Latin America: fraud is growing and the photo is part of it
Data and cases from the region show the same trend as in the UK and the United States.
Brazil: more proven fraud
In 2024, out of R$ 41 billion in claims, R$ 5.4 billion were flagged as suspicious and R$ 1.1 billion were confirmed as fraud.3 Proven fraud rose from 1.8% to 3.3% of earned premiums between 2021 and 2025.12
Argentina: AI already shows up in cases
CESVI flags AI image tampering in claims as a rising trend.7 Adjusters and lawyers are already receiving fabricated photos that pass basic checks.8
The photo arrives by WhatsApp
Many self-inspections and claims are handled by chat, which compresses images and strips much of their metadata. The analysis has to look at the image itself.
7. What’s next
Faking gets cheaper
Each new generation of image tools makes editing faster and more realistic. An analysis of the AI Incident Database concluded that deepfake fraud now operates on an industrial scale, with tools that are cheap and easy to use.15
More investment in detection
The fake image detection market is estimated at USD 1.87 billion in 2026 and could reach USD 7.43 billion by 2031, growing close to 32% a year.16
More labeling, but not everywhere
The EU AI Act requires AI-generated content to be disclosed from August 2026, and watermarking becomes enforceable for tools already on the market in December.1718 It helps, but it doesn’t reach anyone using tools that don’t follow the rule, and a mark can be lost when the image is edited or forwarded.
Photos signed at capture
Some phones already sign photos with the C2PA standard.19 For now few devices do, the signature is often lost when the image is forwarded or cropped, and a photo without a signature proves nothing. It complements analysis; it doesn’t replace it.
8. What to ask a vendor
If you are evaluating a tool to review claim photos, these questions separate a complete solution from a partial one.
Does it analyze the image, its history and the claim’s context?
Each signal alone leaves gaps: a new edited photo won’t show up in any search, and a recycled photo may carry no trace of editing.
Does it explain the reason behind each result?
To deny a claim or refer it for investigation, you need to know why it is suspicious. A number without an explanation doesn’t hold up with the policyholder or the regulator.
Is there a person for the doubtful cases?
Human review of gray-area cases avoids denying honest policyholders and keeps the reasoning on record.
Can you use it without integrating anything?
If the only option is integrating it into the claims system, testing takes months. With a web tool you start in days.
Does it work with self-inspection photos, screenshots and WhatsApp?
That is how real claims arrive: compressed, cropped and missing much of their metadata.
Does it find the same photo in other claims, even if cropped or rotated?
It is the simplest way to find recycled photos and duplicate claims.
Can I test it with my own cases?
A pilot with your real claims shows the results before you sign.
9. Checklist for claims and fraud teams
- Measure what share of your claims is approved with policyholder photos alone.
- Analyze every claim photo, not only the ones that look suspicious.
- Ask for self-inspection photos through the app camera, in the moment, not from the gallery.
- Search every photo online and in your past claims before paying.
- Review the policyholder’s and the policy’s history before approving.
- Define what evidence you need to deny or refer a claim and keep it on record.
- Separate clear cases from doubtful ones, and send only the doubtful ones to an adjuster.
- Train your team with real examples of edited photos.
- Review your rules against the regulation in your market.
How Vouch solves it
Vouch analyzes every claim photo in under 30 seconds with three layers: detection of images generated or edited with AI, a search for the image online and in previous claims, and the policyholder’s context. It returns a 0–100 risk score with the reason, and gray-area cases go to a specialist. It works from Vouch Web, with no integration, or via API.
Frequently asked questions
What is a shallowfake in insurance?
It is a real photo retouched with editing apps or AI to add or exaggerate damage, swap a number plate or reuse an image from another claim. Unlike a deepfake, it isn’t built from scratch, so it is easier to make and harder to notice.
How do you detect a claim photo edited with AI?
It’s getting harder by eye. What works is combining technical image analysis, a search for that same photo online and in past claims, and the policyholder’s context, with human review for doubtful cases.
What tools detect AI-edited photos in insurance claims?
There are general detectors for AI-made images, but for claims you want a tool that combines image analysis, repeated-photo search and the policyholder’s context. Vouch, for example, analyzes every photo in under 30 seconds, returns a 0–100 risk score with the reason and routes doubtful cases to a specialist.
How much insurance fraud is there?
It depends on the market. UK insurers detected £1.34 billion in fraudulent claims in 2025; Brazil confirmed R$ 1.1 billion in 2024; and in the United States it is estimated at USD 308.6 billion a year.
Sources
This report is based on public sources only; every figure can be verified.
- Verisk. AI Editing Tools Are Fueling a New Era of Insurance Fraud, According to New Research from Verisk (Mar 2026)
- Bodyshop Magazine (datos de la ABI). Fraudulent insurance claims worth £1.34bn detected in 2025 (2026)
- Sindicato das Seguradoras. Fraudes em seguros somam R$ 1,1bi: como se proteger? (Aug 2025)
- Coalition Against Insurance Fraud. The Impact of Insurance Fraud on the U.S. Economy (2022)
- PropertyCasualty360. Fraudsters using AI to manipulate images for false claims (May 2024)
- Revista Apólice. Setor de seguros reage a fraudes com inteligência artificial (Apr 2025)
- El Seguro en Acción. Concurso Nacional de Lucha contra el Fraude 2025 (2025)
- Forbes Argentina. El fraude perfecto llegó a los seguros: así están usando IA para inventar daños “creíbles” (Feb 2026)
- Admiral Group. The White Lie Effect: 1 in 8 people admit to exaggerating insurance claims (2026)
- Aviva plc. Aviva stops record levels of claims fraud as scams become more sophisticated (Jun 2026)
- Allianz UK. Posting car pics fuels fraud (Jun 2026)
- Revista Apólice. Fraudes no seguro atingem maior nível da série histórica (Jul 2026)
- Bolsamanía (Europa Press). Axa evita 87 millones en fraude al seguro en 2024, un 3,6% más (2025)
- Insurance Business. Fewer fraud cases, fatter claims as average scam hits £14,300 (2026)
- The Guardian. Deepfake fraud taking place on an industrial scale, study finds (Feb 2026)
- Mordor Intelligence. Fake Image Detection Market Size and Share Research Report, 2031 (2026)
- Jones Walker LLP. The EU approved a high-risk AI delay, but most transparency obligations remain (2026)
- William Fry. EU AI Act Omnibus Deal Reached: Postponed Deadlines, Watermarking Compromise, and the Nudification Prohibition (May 2026)
- Google Security Blog. Pixel and Android: trusted images with C2PA Content Credentials (Sep 2025)