Carpe Vision
Carrier-Submitted Photos Are No Longer a Source of Truth
Carpe Vision uses AI-powered image and document forensics to detect manipulated claim evidence. Eleven detection engines combine visual analysis with claim context, helping carriers focus investigations and reduce fraud-related loss leakage.
AI-Powered Forensics
Detect manipulation and AI-generated content in claim evidence
Carrier Decisions
Claim context and visual findings help focus investigations
Positive Findings
Carpe reviewers check suspected fictitious or AI-generated evidence to help reduce false positives
One Assessment, Every Engine, on the Same Screen
Manipulated and AI-generated claim photos can hide fraud in evidence carriers trust. Carpe Vision runs eleven forensic engines, including three dedicated to AI-generated and AI-edited content, and brings claim context, a risk assessment, confidence score, engine findings, and manipulation heatmap onto one screen. Claims teams can see what raised concern and focus their investigation, with supporting analysis ready for the claim file or SIU referral instead of scattered across separate reports.
Eleven Engines Behind Every Assessment
Eleven forensic engines run in parallel on every image, including three dedicated to AI-generated and AI-edited content. The callouts below show how Vision examines a claim photo for signs of manipulation.
Pixel manipulation analysis
Clone-stamp, splice, and copy-move detection on the damage zone.
Compression artifact analysis
JPEG quantization mismatches around edited regions like the windshield.
Lighting and shadow consistency
Highlight direction on the hood vs. the scene sun angle.
Reverse-image matching
Has this photo, or any region of it, appeared online before?
AI-generation and AI-edit detection
Three engines tuned to synthetic content, including selective AI edits.
Metadata and provenance forensics
EXIF GPS, capture date, device, software, C2PA, against the reported claim.
Pixel manipulation analysis
Clone-stamp, splice, and copy-move detection on the damage zone.
Compression artifact analysis
JPEG quantization mismatches around edited regions like the windshield.
Lighting and shadow consistency
Highlight direction on the hood vs. the scene sun angle.
Reverse-image matching
Has this photo, or any region of it, appeared online before?
AI-generation and AI-edit detection
Three engines tuned to synthetic content, including selective AI edits.
Metadata and provenance forensics
EXIF GPS, capture date, device, software, C2PA, against the reported claim.
A Single Verdict, Backed by Eleven Engines
Pixel manipulation, copy-move, splicing, compression, lighting and shadow, metadata integrity, reverse-image matching, synthetic-media detection, AI-edit detection, document forensics, and provenance signals. One verdict, one confidence score, per-engine evidence underneath.
Claim-Aware Context
Hidden EXIF (GPS, capture date, device, software) is checked against the facts of the claim. A photo captured 200 miles from the reported loss, or three weeks before the reported date, is a fraud signal most tools never see.
Built for the AI Fraud Era
Three of the eleven engines target AI-generated and AI-edited imagery, including the emerging real-photo with selective AI edits category that legacy fraud tools cannot catch.
From Claim Evidence to Informed Action
Bring image forensics into FNOL, SIU referrals, or batch claims review. Vision turns submitted evidence into findings your team can use to prioritize investigation and determine the next step.
01 / Submit
Submit Claim Evidence
Provide claim images and documents with the relevant claim context.
02 / Screen
Automated Screening
Eleven engines assess manipulation and AI generation, using claim context to produce a risk assessment with visual evidence.
03 / Verify
Reviewed by people, not algorithms.
Carpe reviewers verify positive findings for suspected fictitious or AI-generated evidence to help reduce false positives.
04 / Decide
Your Team Decides
Use the findings and supporting evidence to prioritize follow-up and make an informed claim decision.
Where Carpe Vision Earns Its Keep
Vision is built for the loss types where image fraud is hardest to spot manually and most expensive to miss.
Commercial Auto and Heavy Trucking
Staged-collision photos, reused damage shots across multiple claims, and AI-edited vehicle damage are some of the fastest-growing fraud patterns in commercial auto. Vision flags reused images, manipulated regions, and metadata that conflicts with the reported loss location before reserves are set.
Commercial Property
Property loss photos are now routinely altered to expand damage areas, fabricate water lines, or insert synthetic debris. Vision returns pixel-level heatmaps that show exactly which regions of an image have been edited, and compares EXIF location and capture date against the reported event.
Workers Compensation
Comp claims rely on photo evidence of injuries, work environments, and medical equipment. Vision flags reused stock photos, AI-generated injury imagery, and documents that have been digitally altered after capture, with chain-of-custody evidence claims handlers can attach to a file.
DBA and Subcontractor Receipts
Fake receipts, doctored invoices, and overseas subcontractor documents are a known leakage vector. Vision evaluates document layout, font rendering, signature placement, and pixel-level edit traces to surface tampered paperwork before payments go out the door.
Purpose-Built for Insurance Teams
Plug Carpe Vision into the FNOL pipeline, SIU referral workflow, or batch claims review. Vision can score thousands of images at a time through the API, or one image at a time inside an investigator workflow, giving claims teams a faster path to confident, defensible decisions on every photo and document.
Testimonials
What Carriers Say About Carpe
“The alerts are timely, easy to understand, and have already influenced the direction of numerous claim investigations. … Carpe has proven to be a responsive and collaborative business partner. Their team is quick to answer questions, welcomes feedback, and demonstrates a genuine commitment to supporting our success.”
Blaise Schroeder · Vermont Mutual Insurance Group®
Claim Manager
“By integrating new data elements from [Carpe] into our insurance programs for small businesses, we can better understand the challenges faced by business owners and identify options to address their commercial insurance needs.”
Sharon Fernandez · Farmers Insurance
President of Business Insurance
“[Carpe] is the latest addition to Zurich’s ongoing innovation programs focused on transforming the future of insurance…[Carpe] supports Zurich’s ongoing efforts to improve the claimant experience by expediting low-risk claims, and by providing new insights to help our people make informed, accurate, and timely decisions.”
Scott Clayton · Zurich Insurance Company Ltd
Head of Claims Fraud, UK
“[Carpe] has been a great partner in helping us strengthen the claims handling process…Online Injury Alerts helps [us] continue to meet our goals of increased efficiency, reduced potential fraud and improved customer experiences.”
Douglas L. Kratzer · The Hanover Insurance Group
Vice President, Claims
Common Questions About Carpe Vision
What is Carpe Vision?+
Carpe Vision is AI-powered image and document forensics for insurance claims. Eleven detection engines assess submitted evidence for signs of manipulation and AI generation, combining the results with claim context in one risk assessment. Confidence scores and visual findings help claims teams decide what warrants investigation.
Can Carpe Vision detect deepfakes and AI-generated images?+
Yes. Three of the eleven forensic engines are dedicated to AI-generated and AI-manipulated imagery, including the harder "real photo with selective AI edits" pattern that most fraud tools miss. Each engine is retrained continuously against the current generation of synthetic-media tools.
How does Carpe Vision use claim context?+
Vision compares hidden image metadata (GPS coordinates, capture timestamp, capture device, editing software) against the facts the claimant provided (reported date, reported location, claimant identity). Mismatched GPS, dates, or devices are some of the strongest fraud signals available, and Vision surfaces them as structured flags with the source evidence attached.
Does every file receive human review?+
No. Human verification applies to positive screening findings for suspected fictitious or AI-generated evidence, with Carpe reviewers assessing whether the concern is supported to help reduce false positives while the carrier retains responsibility for claim decisions.
How does Carpe Vision fit into a carrier workflow?+
Vision supports API-based screening at FNOL, SIU referrals, and batch claims review, as well as individual-image analysis for investigators. Its risk assessments and supporting evidence give teams a consistent basis for prioritizing follow-up within their existing claims process.
How does Carpe Vision fit with the rest of the Carpe suite?+
Vision can stand alone for any carrier, MGA, or platform that wants image and document forensics, or it can plug into Carpe Case Management for an end-to-end claims fraud workflow alongside Online Injury Alerts, Investigative Reports, and native open-source intelligence.
Why is Carpe Vision necessary for carriers?+
Generative AI has collapsed the cost of producing convincing fake imagery. A claimant with a phone can now insert damage that never happened, alter a date stamp, fabricate a receipt, or generate a synthetic injury photo in seconds, and most legacy fraud tools only look at pixels in isolation. The result is loss leakage hidden inside the photo evidence carriers already trust. Carpe Vision is built for that environment: eleven forensic engines run in parallel on every image, three of them dedicated specifically to AI-generated and AI-edited content, with visual evidence to explain the automated findings.
See Carpe Vision on Your Own Images
Run Vision on a sample of your claim images, explore the findings and visual evidence, and see how it fits your FNOL, SIU referral, or batch review workflow.