PhotoDetective – Detecting AI Generated and Internet Sourced Fraud Qorus-NTT DATA Innovation in Insurance Awards 2026

PhotoDetective introduces advanced AI‑based detection tools into PZU’s claims processes, enabling the identification of image fraud such as AI‑generated visuals, digitally altered photos, and pictures copied from the Internet. The initiative was launched due to the rapid development of generative AI, which makes it extremely easy to create realistic fake damage photos or modify real ones. PZU previously lacked any tools capable of detecting this type of fraud, making it impossible to assess the scale of the problem and exposing the organization to financial risk.

The project was inspired by the growing availability of generative tools, real cases of fraud detected manually, and the urgent need to develop internal deepfake‑detection competencies. During the preparation phase, ten vendors were analyzed; four responded, and two were qualified for the pilot phase. For the pilot phase, a dataset of 5,000 images was created, consisting of verified real damage photos and a sample (approximately 20%) of images deliberately generated or manipulated using AI tools, as well as images downloaded from the Internet.

The project involved: Claims Department, Insurance Fraud Prevention Unit, Innovation & AI Team. Pilot results showed high effectiveness — 81% accuracy on manipulated images and a very low false‑positive rate (<1%), validated through manual reviews. The project also confirmed that image fraud is a real problem and that human operators find it extremely difficult to distinguish real images from deepfakes.