New analysis from SAS shows how generative AI can add believable “damage” to everyday photos – a tactic fraud investigators and claims handlers say is getting harder to spot as the tools spread.
The challenge is simple: two of the three images are fabricated or altered, and one is genuine. Before scrolling, decide which you would accept as authentic evidence in a claims file.
The set shows a coffee table with what appears to be a small crack (pictured above), an armchair with what looks like a coffee stain, and a car interior with glass smashed into the vehicle (both below). SAS said 40% of people cannot tell which one is real.
SAS said its work demonstrates how AI can fabricate crash scenes, property damage and other supporting evidence in seconds, reflecting methods it said are being used to deceive insurers. It linked the issue to both individual fraud attempts and organised crime activity.
The company also pointed to “vanilla synthetics” - small edits that can be difficult to detect once attached to a claim - and warned such manipulations may pass a quick human review, particularly when the change resembles routine wear-and-tear.
The Insurance Fraud Register has said insurance fraud has led to an average increase of £50 on annual premiums for consumer policies. Adyen has reported the average cost of a fake claim has hit £84,000, and that one in seven claims is proven to be fraudulent.
“Fraudsters are exploiting generative AI tools to make fabricated damage and doctored scenes look entirely plausible,” said Adam Hall, an insurance specialist at SAS. “With just a few prompts, they can create, enhance or erase visual evidence to support a false insurance claim.”
Hall said assessors should look for “subtle inconsistencies”, including shadows that fall the wrong way, damage that does not match the impact, blurred number plates, or backgrounds that appear too clean or empty. “These tiny visual mismatches are often the first red flags of an AI-generated claim,” he said.
SAS said the coffee table crack and the armchair stain were AI-doctored. It said the smashed glass inside the vehicle was the real image.
Hall said AI is also being used by insurers to identify suspicious activity at scale, including patterns linked to repeat behaviour and organised networks.
“AI and machine learning can detect both one-off scams and sophisticated, organised networks,” he said - tools he argued can help insurers triage claims risk and reduce fraud losses.
