SAS says generative AI is fueling synthetic claims fraud

SAS is treating synthetic claims evidence as the next fraud arms race. The company said generative AI has made it startlingly easy to create or alter realistic-looking images for insurance claims, turning what used to be manual photo tampering into a fast, scalable attack on claims intake.

The pressure showed up in a customer request at SAS Innovate, where a carrier came in after seeing a surge in fraudulent receipt images. SAS ed with an agentic fraud-screening pipeline built on SAS Intelligent Decisioning, pairing computer vision, OCR and large language model reasoning to flag synthetically generated or manipulated images before they influence a claims decision.

That matters because the workflow problem is bigger than one bad file. Claims teams now have to inspect pixels, text and surrounding context together, not rely on a rule that checks whether a document looks obvious or whether a receipt number matches a pattern. If image authenticity stays a manual review problem, carriers will miss the volume game. Auto and property claims are especially exposed, since photos are central to the file and adjusters are already juggling speed, severity and customer expectations. The practical move is to push verification into the claims platform itself, early enough to stop leakage without turning every honest customer into a fraud suspect.

The scale of the risk is why insurers are paying attention. The FBI says insurance fraud costs roughly $30 billion a year and adds about $200 to $300 per family in higher premiums. The National Insurance Crime Bureau later put the annual hit at about $308.6 billion, or roughly $900 more per policyholder, and said its analysis of thousands of questionable claims from 2022 through June 30, 2025 showed a significant year-over-year increase in claims involving identity theft or synthetic identity. Nearly a quarter of claims referred for identity-theft reasons involved a synthetically generated identity.

Industry warnings are lining up behind that picture. Guidewire said insurers are now facing deepfakes, shallow fakes and synthetic media, and that traditional review methods and legacy fraud models were never built to catch these manipulations. Debevoise & Plimpton added that AI image manipulation can fabricate or exaggerate claims and can distort reserve estimation if detection lags.

SAS’s point is straightforward: fraud controls cannot stop at another rule. Carriers need a verification layer that can score image authenticity, read the text inside the file, and weigh the narrative around it before a claim moves deeper into the workflow. That is how the next generation of claims fraud gets contained.