As an AI company, it may seem unusual for us to talk about the dangers of AI, but unfortunately, that’s the reality. It’s a double-edged sword. People are using AI to generate fake images of damaged property to fraudulently claim compensation. We are preventing this by using AI to analyze images taken in real time, ensuring they haven’t been modified.
To put it simply, we’re preventing fraud by AI with AI.
But how realistic are AI generated images really?
To put AI image generation to the test, we asked ChatGPT to “generate a realistic photo of a BMW X5.” Here’s the result:
Next, we wanted to add some damage, so we prompted: “add some scratches to the front bumper.” The result was surprisingly convincing:
To further manipulate the image, we asked ChatGPT to change the license plate to read “123 ABC” instead of “BMW X5.” Here’s the updated photo:
At this point, the image already looked quite realistic, but we wanted to refine it further. To make it appear more authentic, we reduced the image quality and formatted it to match typical phone camera dimensions. Here’s the final result:
Upon closer inspection, the image appears a bit too polished to be a real photograph. However, when viewed casually, such as when scrolling through a collection of photos, a claims handler might not immediately notice the inconsistencies.
Now, this wasn’t a tutorial on how to commit insurance fraud, but rather a demonstration of how little effort it takes to generate a fake image. If we added some Photoshop on top of that, making the car body a bit dirtier and adding some dirt on the ground for example, it would be quite difficult to tell whether the image was real or not.
How much fraud is there in motor insurance?
The Coalition Against Insurance Fraud (an organization dedicated to educating and protecting American citizens from the cost and damage of insurance fraud) published a study in 2022 to develop a new estimate of the economic impact of insurance fraud in the US.
The study referenced a statement from the Insurance Information Institute, which states on its website:
“In the late 1980s, the Insurance Information Institute interviewed claims adjusters and concluded that fraud accounted for about 10 percent of the property/casualty insurance industry’s incurred losses and loss adjustment expenses each year.”
Based on this, the study assessed the cost of P&C fraud from 2016 to 2020 by taking the loss adjustment expense (LAE) for each year and then multiplying that figure by 10%. Here’s the result:
According to Statista, in 2023 private passenger auto insurance premiums accounted for approximately 36% of all P&C insurance premiums written in the US. Since auto insurance is the largest segment of the P&C insurance market, it also contributes the most to overall P&C fraud.
It’s important to note that these figures are based on data from a few years ago and do not account for the recent advancements in AI, which could have significantly impacted fraud detection and prevention. Without more recent studies, the true extent of fraud today remains uncertain. We’ll have to wait and see.
The solution
As fraudsters now have thousands of AI tools at their disposal, insurance companies must also equip themselves in response. While insurers could develop their own AI solutions, we can assure you, it’s no picnic. In our experience, it’s far more efficient for insurers to focus on what they do best and outsource the necessary solutions.
For outsourcing to be as seamless as possible, AI add-ons should be lightweight, easy to use, and integrate smoothly into existing processes. With DriveX, that’s exactly what we aim for. We couldn’t wrap up this blog post without a little promo now, could we? 😉 But seriously, from an insurer’s perspective, eliminating one source of fraud – faked vehicle photos – is easy with DriveX.