The new face of insurance fraud? How AI-doctored images are becoming more realistic
By Adam Hall, Insurance Fraud Specialist, SAS
Generative AI tools and deepfake apps have made image manipulation accessible to almost anyone. While some platforms have become wise to potentially criminal behaviour - refusing to generate images they suspect could be used for fraud - others still allow users to create manipulated visuals using artificial intelligence with far fewer safeguards.
According to the Insurance Fraud Register, insurance fraud has now led to an average increase of £50 on annual premiums for consumer policies - while the average cost of a fake claim has now hit £84,000, with one in seven claims proven to be fraudulent, according to global fintech platform Adyen.
Bad actors fabricate damage, erase incriminating evidence, or construct entirely fake scenes in minutes. With image-generation tools capable of producing highly realistic visuals at speed, the industry is facing urgent questions about the impact on claims processes and evidence verification.
It has become alarmingly easy to make a car appear damaged or a home look vandalised. And because criminals face no regulatory constraints, they’re becoming more inventive in how they exploit these tools.
And it’s not just a few rogue chancers. Organised Crime Groups (OCGs) are increasingly polycriminal, engaging in a wide range of illicit activities. Many are adopting generative AI at scale as part of these broader criminal networks. Insurers, meanwhile, are spending hundreds of thousands of pounds trying to detect doctored images. Those that are digitally altered, blended, or fully synthetic content have become so convincing that distinguishing genuine evidence from AI-manipulated material is increasingly difficult.
How image doctoring fraud works
Fraudsters are exploiting generative AI tools to make fabricated damage and doctored scenes look entirely plausible. With just a few prompts, they can create, enhance or erase visual evidence to support a false insurance claim.
Adding fake damage
Car insurance claims are one of the most prolific for insurance fraud. Using simple prompts such as “Make my old car look like it hit a tree” or “Add dents and broken glass to the front bumper”, fraudsters can instruct image-generation platforms to create highly realistic crash damage. These tools can:
- Add scratches, dents, shattered windscreens or deployed airbags
- Produce multiple “angles” of the same fabricated scene
- Adjust lighting and reflections to make the damage appear authentic
- Match the vehicle’s colour, make and model with near-perfect accuracy
Some are using these doctored images to inflate genuine claims, while others fabricate entire incidents that never occurred.
Removing incriminating details
AI editing tools also make it straightforward to delete elements that could contradict a claim. Fraudsters can seek to erase:
- Eyewitnesses in the background
- Number plates
- CCTV cameras
- Surrounding vehicles
- Skid marks or debris inconsistent with their story
This allows them to remove data points that insurers would normally analyse to reconstruct what actually happened. By stripping away context, they reduce the chance of being caught out by inconsistencies.
Constructing entirely fake scenes
More sophisticated image generators can build full accident scenes from scratch - complete with environmental details like road markings, weather conditions, shadows and reflections. Criminals can create:
- Non-existent collision sites
- Staged burglary scenes
- Property damage that never occurred
Organised crime groups often use these tools to mass-produce convincing imagery across multiple fraudulent claims
Blending the real with the synthetic
One of the most challenging trends for insurers is the rise of “hybrid” images - real photos subtly enhanced with AI. Fraudsters might take a legitimate photo of a scratched bumper, then amplify the damage using AI editing tools. Because the photo is partially real, it becomes even harder for claims handlers to detect tampering.
Adam Hall
Insurance Fraud Specialist, SAS
SAS’ test: Can you spot how these images have been doctored?
Altered image 1 - Realistic collision
This image is entirely AI-generated - a demonstration of just how convincing synthetic visuals have become. Even with the number plates intentionally blurred, a simple prompt such as “Create an image of two Hondas - a blue Civic and a red Accord - in a collision on a suburban English street” produces a scene that appears realistic and, to the average viewer, completely believable.
Synthetic image 2 - Remote locations
One tactic increasingly used by OCGs is removing any contextual clues that could help insurers to verify whether an incident actually occurred. By erasing eyewitnesses, passing vehicles, road signs or environmental inconsistencies, fraudsters can create a scene with no external reference points - making it far harder for investigators to cross-check details.
In this example, the prompt “Create an image of a white van crashed into a tree in the English countryside with nobody around” produces an isolated, barren scene. With no bystanders, traffic, reflections, shadows from other objects, or identifiable surroundings, insurers have very little to analyse.
Synthetic image 3 - Closer up
We asked AI to ‘Create another image closer up of the van crashed into the tree’ erasing even more potential evidence. This kind of synthetic “clean” image can strip away the usual signals that claims teams rely on - such as witness statements, background movement, or location-specific markers - allowing fraudulent claims to slip through undetected unless specialised forensic analysis is used.
Amended images
Not all fraudulent images are fully AI-generated - in many cases, genuine photos are subtly amended using generative AI to add, remove or alter details. And it isn’t limited to cars; fraudsters are manipulating images of receipts, properties, holiday rentals and more.
A recent case involving an Airbnb stay shows how serious this can be. One guest was initially billed more than £12,000 for alleged damage to a New York apartment, including a stained mattress and a cracked coffee table. However, she challenged the claim, arguing the host had digitally altered the photos. Airbnb later apologised, issuing her a refund of nearly £4,300 after finding the images had indeed been manipulated.
Removed bystanders
In this example, the prompt “Remove all people from this image, change the registration plates of the cars behind, and add damage to the yellow car’s windscreen” produces a scene that looks far more credible than it should.
By stripping away bystanders and altering surrounding vehicles, the image removes key pieces of contextual evidence that insurers typically rely on. With the environment sanitised and the damage artificially added, validating whether the collision actually occurred becomes significantly more difficult.
Campervan registration plate changes
Prompt: “Change the registration plate and create damage on the front of the van that looks like someone has driven out in front of it, caused it to brake hard, then driven off.”
This kind of manipulation once again strips away crucial evidence. By altering the number plate and adding realistic-looking damage, the image becomes far harder for insurers to interrogate. The scene appears plausible, yet gives no reliable context - especially as the campervan is depicted in a remote location where “anything could have happened.” The result is a lifelike but unverifiable image that can easily be used to support a fabricated claim.
Decor changes
As seen in the recent Airbnb case, image manipulation isn’t limited to vehicles. Fraudsters can alter photos of rental properties, home décor or household items to claim money back from tenants, recover deposits, or even file warranty or insurance claims.
In this example, the prompt “Add a small but noticeable crack to the glass coffee table” produces a subtle yet convincing defect. The alteration is minimal but looks entirely authentic - exactly the kind of damage that could be used to support a false claim. These small, realistic edits are particularly dangerous because they’re harder to spot and often slip through without raising suspicion.
Fake stains
Similar to décor edits, fraudsters can fabricate small but convincing stains to support false insurance or deposit claims. These subtle alterations are easy to generate and difficult to dispute - especially when they appear on everyday items.
In this case, the prompt “Add a coffee-like stain onto the chair” produces a realistic mark that could easily be passed off as genuine damage. It’s a simple edit, but one that can have significant financial implications if used dishonestly.
How can insurers prevent insurance fraud?
While AI can be used to commit insurance fraud, it’s also a powerful tool to detect it.
AI and machine learning help insurers spot both simple “one-off” fraud and complex, organised crime scams. By analysing vast volumes of claims and related data, AI detects anomalies and patterns that human analysts might miss - improving accuracy, reducing losses, and protecting profitability.
Modern AI platforms aggregate internal and external data - policyholder information, claims history, public records, device data - into a 360-degree view. This creates a complete picture of each claim and streamlines workflows by integrating data management, alerting, case handling, and decision-making.
Through social network analysis, AI uncovers connections between seemingly unrelated claims, individuals, and organisations, exposing coordinated fraud networks. AI models also reduce false positives, letting human investigators focus on the most likely fraudulent cases, improving efficiency and lowering costs.
As fraudsters adopt new techniques - fake identities, forged documents, digital-first scams - AI evolves too, retraining models, integrating new data sources, and updating risk scoring. This can keep insurers ahead of emerging threats.
ENDS
Methodology
SAS leveraged generative AI to simulate and analyse multiple scenarios of common insurance fraud techniques. By issuing a series of tailored prompts, the AI generated examples and insights across different types of fraudulent activity.
This approach allowed SAS to highlight how easy it can be to manipulate images, providing a comprehensive view of the key challenges facing the insurance industry today. Where existing images were doctored, they were taken from free-to-use image sites.