The photo was a real camera shot. Not a deepfake. That's exactly why it nearly slipped through, it was lifted from a claim that already settled. That's the fraud almost nobody builds for. Everyone is racing to spot AI-generated images, but a genuine photo pulled from an old, settled claim passes every authenticity test, because it really is genuine. A deepfake detector waves it straight through. When we built this demo, a forensic authentication gate for auto claims, we watched it catch exactly that. The image passed as authentic, because it was. What flagged it was a perceptual-hash match against the historical claims index, the same picture had turned up in an earlier claim, alongside a narrative that didn't match what the photo actually showed. Reused evidence, routed to the fraud unit with a full record of why. (The claim, the photo, and the past-claims index are all synthetic, built for the demo, so you can watch a catch happen without anyone's real file on the line.) On a labeled 120-claim synthetic set, a single-signal authenticity detector caught 0 of the 35 recycled and metadata-spoofed frauds. Cross-signal forensics caught all 35. Real photos, quietly reused, are the gap. If you work claims or fraud: when a submitted photo comes back genuine, does anything check whether it is genuinely new, or does "real" end the inquiry? #InsuranceFraud #ClaimsManagement #SIU #InsurTech