Industries / Finance and business services / Insurance
Fighting fraud
Finding claims and applications that are invented, staged or inflated, and the people behind them.
5
Milestones achieved, from 1912 to 2025.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1912
Insurers pool their detectives
After three Chalmers cars were stolen in Chicago in the summer of 1912, Fred J. Sauter led insurers in forming the Automobile Protective and Information Bureau, so that they could chase stolen cars together rather than one by one. Its successors merged in 1992 with the Insurance Crime Prevention Institute to form the National Insurance Crime Bureau (NICB), which is supported by more than 1,100 property and casualty insurers and self-insured organizations.
Credit
Fred J. Sauter
2006
A bureau for staged crashes
UK insurers set up the Insurance Fraud Bureau (IFB) in 2006, with five staff, to lead the response to organized motor fraud, especially "crash for cash", in which criminals stage collisions to claim. By July 2016 its work had contributed to 1,181 arrests and 442 convictions.
Credit
Insurance Fraud Bureau
2021
An AI claim withdrawn
In May 2021 the US insurer Lemonade posted that its AI could "pick up non-verbal clues" in customers' claim videos. Critics objected, and Lemonade replied that its systems do not judge claims on "background, gender, appearance, skin tone" and similar traits. Within days it deleted the thread, which it called "awful", and said AI is never used to automatically deny claims.
LessonCustomers and regulators expect an insurer to say exactly what its fraud tools look at and who makes the final decision.
Credit
Lemonade, and the critics who questioned it
2022
A price on fraud
A study for the Coalition Against Insurance Fraud put the cost of insurance fraud in the US at $308.6 billion a year, across all lines of insurance. It compiled figures from the FBI, the GAO, CMS, which runs Medicare and Medicaid, and others. The coalition's 1995 figure of $80 billion covered fewer lines, so the two cannot be compared directly.
Credit
Coalition Against Insurance Fraud; Colorado State University Global's White Collar Crime Task Force
2025
Fewer frauds, larger sums
UK insurers detected 93,900 fraudulent claims in 2025, 2.7% fewer than in 2024, worth £1.34 billion, 14% more, by the Association of British Insurers' count. The average was about £14,300. Motor made up 55% of the claims, and contrived incidents, those staged or caused on purpose, grew fastest, by 30%.
Credit
Association of British Insurers (ABI), as reported by Insurance Times
Watch for
Photos proven at capture
Truepic says its Truepic Vision checks each claim photo and video at the moment it is taken and runs it through more than 50 fraud checks, and that carriers using Verisk's ClaimSearch, a claims database shared across insurers, can now request such photos from within their workflow. The need is measured: in a Verisk survey of 300 claims professionals from December 2025 to January 2026, only 32% said they were very confident at spotting deepfakes, and 36% of 1,000 consumers said they would consider editing claim photos or documents.
Achieved whenan insurer or fraud bureau publishes measured results, such as altered images caught and false alarms, for claims handled this way.
Claimed by
Truepic and Verisk, September 2026
Open
01
Doctored photos for testing
A public set of car and home damage photos, half untouched and half altered with current image-editing models (an added dent, an enlarged water stain, a swapped number plate), with a mask showing what was changed in each. It scores detection tools on what they catch and what they wrongly flag.
The briefCredit · unclaimed02
A ghost-broker checker
A free tool where a driver pastes a social media post, message or web address offering car insurance. It pulls out the firm's name, phone number and website and checks them against the FCA's register of authorized firms, including whether the contact details match, then gives a plain verdict with its reasons.
The briefCredit · unclaimed
Open 01 · the brief
Doctored photos for testing
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of car and home damage photos, half untouched and half altered with current image-editing models (an added dent, an enlarged water stain, a swapped number plate), with a mask showing what was changed in each. It scores detection tools on what they catch and what they wrongly flag.
Why it is still open
In Verisk's 2026 survey, 38% of claims professionals said their tools miss too many altered images and documents. The closest published work, a 2025 paper presenting the vehicle-inspection company UVeye's approach, describes a product and releases no data. Public forgery sets found in this search cover receipts and documents, not damage photos.
Built with what exists
CarDD images, used under its authors' license; current image-editing models from Google, OpenAI and open-weight developers; open-source image forensics models as the first detectors to score. The images are for testing detectors only and are never submitted as claims.
First step
Take 100 CarDD images, make 100 edited copies with one editing model, and record each edit.
Done looks like
2,000 images, 1,000 of them altered, each with an edit mask. At least three detectors scored on detection rate and false alarms, and a held-back set of 500 images so that scores cannot be tuned to the test.
Open 02 · the brief
A ghost-broker checker
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A free tool where a driver pastes a social media post, message or web address offering car insurance. It pulls out the firm's name, phone number and website and checks them against the FCA's register of authorized firms, including whether the contact details match, then gives a plain verdict with its reasons.
Why it is still open
Ghost brokers sell fake or invalid car insurance through social media and messaging apps. In FCA research with 1,000 UK drivers aged 17 to 25 (April and May 2026), 49% had bought insurance through social media or messaging apps and 39% were not confident they could spot a fake policy. The FCA's advice is a manual check with its Firm Checker.
Built with what exists
the FCA Financial Services Register and Firm Checker; domain registration lookups; a language model to read the post and extract names, numbers and links.
First step
Collect 50 insurance-related entries from the FCA's warning list and record what details each gave: name, phone number and website.
Done looks like
On a test set of 200 adverts, 100 rebuilt from FCA warnings, police releases and court cases and 100 from authorized firms, the tool flags at least 95 of the fakes and wrongly flags no more than 5 genuine ones. Twenty first-time users each complete a check in under a minute.