Fraudsters flip to AI and deepfakes to make dodgy insurance coverage claims
- Rise of AI is leading to a growing wave of criminals trying to scam insurers
- Cost of this trickery is passed on to honest drivers, homeowners and more
Growing numbers of Britons are using doctored images and documents to make fraudulent insurance claims – with many impossible to spot with the naked eye, This is Money can reveal.
The average insurance fraud was worth £15,000 in 2023, with insurers detecting £1.1billion of fake claims, according to the Association of British Insurers.
Every fraudulent claim pushes up the cost of insurance for honest households, and an emerging technique is to use digitally-altered photos and documents to try to scam money out of insurers.
This can take the form of photos of cars that are doctored to look like they have had a serious crash, or pictures of house contents that have been altered to appear damaged, according to insurance risk and data firm Verisk,
Some of the photos below, supplied to This is Money exclusively by Verisk, show how convincing these fake images can be.
Spot the difference?: In an example of how convincing fake photos can be, the real photo of the car on the left was edited using AI to have damage marks that could deceive an insurer
Similar techniques are used on insurance documents, including insurance policies and no claims discounts for motor insurance, which are being tweaked using computers to try to defraud insurers.
This digital trickery started gathering pace in 2017, according to Verisk product manager, anti-fraud, Kaye Sydenham.
The earliest cases saw fraudsters make fake insurance claims using photos they found online.
But digital photo manipulation has only gathered pace since then, for several reasons.
One is that technology has improved so that very convincing photos and documents can be digitally altered at home – often with free programs and AI software.
Another is that the cost of living crisis has left many people poorer, and therefore with more incentive to commit fraud.
Bureau of Investigation: Again, free AI software quickly made the real photo on the left appear to have scorch marks from a non-existent fire – enough to catch out an unwary insurer
But many insurance frauds are not carried out by those struggling financially, but instead by criminals, often with links to organised crime, who see insurers as easy targets.
That has worsened since the pandemic, when many insurers adopted more remote working patterns and automated claims systems.
Many insurers use online, automated systems for low-value claims, and this is an area ripe for deepfake trickery with claims, Sydenham said.
Low-value claims are often put through automated systems. The exact definition of ‘low value’ varies between insurers, but is £2,000 or lower as a rule of thumb.
Taking humans out of the system makes it easier for fraudsters to sneak a fake claim through.
Canvassing opinion: Would you be able to spot that the roof damage in the right-hand photo is fake, entirely created by software? The real photo of the roof is on the left
The exact number of digitally-altered claims is unknown, but Sydenham says one in 10 of the lower-value claims analysed by Verisk has a red flag for this type of fraud.
Although home and motor insurance sees the most image fraud, other areas of insurance are prone to it too.
‘Any line of business is open to this’, Sydenham said. ‘It is more prevalent in motor and domestic property claims, as they are lower value. But I think it will be rife in travel and pet insurance.
‘AI videos could be coming, such as dashcam videos of fake crashes. It could be a bit of a whack-a-mole, chasing fraudsters and what they might do next.’
Verisk has developed clever technology that can help insurers spot this sort of digital image manipulation.
This software can spot when images have been altered, and helpfully highlight any changes.
It also monitors the metadata of photos – the underlying data that explains how and where a photo has been made.
Verisk can check this metadata for any tell-tale signs that a photo might not be genuine, such as a location that does not match the rest of the claim details.
The software can also perform a similar fraud-busting task with documents, for example by tracking any changes that have been made.
It can even unlock changes that have been made to documents step by step, letting insurers work out exactly where the attempted fraud is happening.
