Fraud in the age of Gen AI: New threats, new defences

17 April 2025 / 4 min read

Generative AI has made it easy for fraudsters to create fake claims with realistic documents, invoices, and images. But the technology is already becoming available for insurers to defend themselves and their customers against this threat.

As generative AI (gen AI)—artificial intelligence technologies that create new content based on their training data and inputs like natural language prompts—explodes into the mainstream, all industries are exploring its potential impact. For insurers, gen AI offers opportunities in automation and efficiency but also unprecedented fraud risks.

Gen AI is transforming the fraud landscape, empowering fraudsters with new tools as well as insights into insurers’ fraud detections. To protect clients and their bottom line, insurers must move quickly to catch gen AI-driven fraud, as it threatens to impact claims costs, premiums, and customers’ trust.

deep-fake-facial-analysis

How gen AI fuels fraud

Gen AI can enable the rapid creation of fake documents such as medical reports, invoices, and accident photos. Fraudsters can generate convincing documents to support their claims in seconds and with little technical knowledge.

“Ask a chatbot to write a medical report for a slip-and-fall, and it’ll create a realistic document with all the right terminology. Fraudsters can get an inside view of what insurers look for in claims,” says Kaye Sydenham, product manager for anti-fraud, Claims UK at Verisk.

These AI-generated documents can deceive human reviewers and existing automated systems, especially when paired with small edits to genuine documents, known as ‘shallowfakes.’ While major frauds are often flagged, many minor adjustments slip through undetected and can cost insurers significantly.

Beyond document creation, gen AI can also give fraudsters deeper insight into how insurers combat fraud. Neil Jones, head of the Claims Investigation Unit at Verisk, warns that fraudsters can use a variety of AI tools, testing how insurers respond to each and finding gaps to exploit.

“Fraudsters may do a full scan of multiple insurers to determine who’s the weakest link. It’ll be those insurers that are targeted,” Jones says.

While writing this article, we tested ChatGPT to see how it could facilitate a fraudulent claim. Asked what defences insurers use to identify fraud, the chatbot gave extensive information covering document and image analysis, data sources, and pattern recognition, giving more details when prompted. It was not until we asked how to get through those defences that it refused to provide more information: “I’m sorry, but I can’t assist with that. However, I’m happy to help if you have questions about legitimate insurance processes.”

Evidently, while some technology companies have introduced barriers to prevent blatant misuse of their products, there are genuine reasons someone may wish to access information about insurance fraud. Thus, gen AI-driven fraud is not an issue to be tackled by technology companies alone; rather, it takes effort from the insurance industry and solutions providers.

AI-driven solutions for fraud detection

To counter the threats posed by gen AI, Verisk and other providers are developing models to spot fake documents and claims.

Verisk's fraud detection tools provide a layered system of defences, designed to catch fraudulent claims from multiple angles. Think of it like a series of filters, each designed to spot a different red flag, increasing the likelihood that a fraudster will be caught on one or more count.

Image analysis detects changes in the noise pattern of photos, providing a heatmap of where edits may have occurred.
Document forensics spot irregularities in PDFs, such as mismatched fonts, altered dates, or overlapping text, as well as whether the document has been used in a different claim.
Deepfake detection identifies wholly AI-generated images, with algorithms trained to spot disparities between computer-generated and real photos.
Virgin identity identification flags profiles with no insurance history, indicating that it may be a false identity trying to evade detection.

Many of our tools rely on similar technology to the gen AI models they counter. They were trained with machine learning, improving iteratively before hitting the market. Now that they are live, our customer feedback mechanism helps Verisk’s experts continue to adapt as new threats emerge.

Looking ahead

Many insurers view fraud as a game of whack-a-mole, with fraudsters’ techniques constantly evolving. Partnering with a solutions provider allows insurers to step out of the game, handing the mallet to an expert whose job is to learn and adapt.

“It’s a moving target. Once you detect document fraud, they’ll try deepfake videos. We’re already preparing for that next step,” Sydenham says.

The ongoing battle against gen AI-powered fraud may seem daunting, but the long history of the insurance industry is one of adaptation and renewal.

“This is the latest ‘next big thing,’” Jones says, “but by working together and embracing new tools, there’s no reason we can’t combat it.”

Stay ahead of AI-driven fraud with our cutting-edge solutions—see how our fraud solutions can help you to protect your business and clients.

Learn more