On 8 October, Laird joined CHO members for a webinar examining the growing challenge presented by AI-generated and exaggerated vehicle damage images.
Led by CHO’s Matt Rowland and Laird’s Nik Ellis, the session included an interactive review of genuine, altered and AI-generated photographs. Attendees were asked to decide which images were real, and the mixed responses demonstrated just how difficult that judgment can be.
The central message was clear: no single visual anomaly proves that an image is fraudulent. The aim is to identify several warning signs that, taken together, justify further investigation.
What should claims handlers look for?
Lighting is a useful starting point. Reflections should remain consistent across adjacent panels, while shadows, weather and daylight should fit the reported time and location.
The damage must also make physical sense. Its height, direction and severity should be compatible with the reported impact, the materials involved and the damage to the other vehicle.
During the webinar, examples included distorted body geometry, unusual wheel positions, inconsistent grazing, repeated image patterns and plastic bumpers appearing to fracture in an implausible way. Number plates, vehicle badges and other text can provide further clues because generated lettering may still contain subtle errors.
A large disparity between the two vehicles is another reason to ask questions. There may be a legitimate explanation, but severe damage to one vehicle accompanied by little more than a scuff to the other warrants closer examination.
Ask for the originals
Image provenance was another important theme. Where concerns arise, claims handlers should request the original photographs and multiple views of the vehicle. The images should show consistent damage and support the account given.
Original files may also contain information about the device, date, time, location and editing software. Missing metadata is not proof of manipulation: claims portals, messaging services and ordinary image processing can remove it. It should nevertheless prompt the question of what happened to the original file. Why was it removed? Innocently by software or is there a more nefarious reason.
The same caution applies to repair estimates and invoices. A PDF can be altered very easily, so checking a document with its originator or obtaining the underlying data directly may be more reliable.
Escalate rather than assume
The webinar repeatedly distinguished suspicion from proof. Two or three meaningful red flags should trigger proportionate escalation, not an immediate allegation of fraud.
Technical examination can assess metadata, cryptographic provenance, pixel manipulation and repeated or moved elements. Where doubts remain, a physical inspection by an experienced vehicle engineer may provide the clearest answer.
An engineer can establish whether the damage exists, whether the materials have behaved as claimed and whether the overall damage profile is consistent with the reported accident. Further images from the repairer or salvage yard can provide additional independent evidence.
AI has lowered the barrier to convincing image manipulation, but the response remains familiar: preserve the evidence, test the account against engineering principles and investigate inconsistencies properly.
When the image does not tell the whole story, informed human judgment still matters.
