98%
of insurers say AI editing tools are fueling digital media fraud
Verisk State of Insurance Fraud Study, 2026ARP
Audit data controlling policy.
Audit Reality Portal is a forensic evidence stack for insurance carriers: live digital provenance, geospatial forensics, and control.
Industry signal · 2026
Media provenance, lack thereof, creating the new age for digital forensics.
89%
of homeowners third-party liability fraud-detection models are vendor-sourced
NAIC AI/ML surveys, 2021–2025 · JIR$1T
in revenue tied to unmonitored AI by 2028—outpacing model accountability and driving drift
Gartner forecast · unmonitored AI, 202866%
believe digital media fraud goes undetected often or very often
Verisk · 300 claims professionals surveyedAdditional context: Coalition Against Insurance Fraud estimates U.S. insurance fraud losses exceed $308 billion annually; FBI/NICB estimates put fraudulent P&C claims near 10% of volume (~$45B). Undetected fakes do not only pay out—they enter loss experience and training data, so automation extrapolates the error. ARP measures stack intake and audits model drift so both the claim and the model can be sanitized.
What carriers are seeing
Fraudulent interpolation is now a claims-and-model problem.
Media enters as First Notice of Loss. Automation accelerates intake. If detection fails, a fictitious claim can be paid—and the same file can unduly extrapolate the next model. ARP scores provenance, compression artifacts, and cross-claim consistency so SIU gets a defensible packet, and contaminated data can be sandboxed out of the internal AI stack.
Fabricated damage stills
AI-generated or inflated vehicle and property damage submitted at FNOL.
Synthetic documents
Repair estimates, receipts, and reports authored by language models.
Identity & voice clones
Caller verification and recorded statements spoofed with cloned audio.
Metadata laundering & recycled images
EXIF stripping, geotag conflicts, re-encoded “originals,” and search-recycled stills.
Live briefing · recent coverage
Tracked articles your risk committee can cite.
NICB: AI editing tools fuel a new era of insurance fraud
NICB coverage of Verisk research: 36% of consumers would consider digitally altering claim images or documents; 98% of insurers link AI tools to rising digital fraud.
Read NICB releaseNAIC: Third-party models sit inside fraud detection
Across NAIC AI/ML surveys, homeowners insurers rely on vendor models more than auto writers for third-party liability fraud detection (89% vs. 60%)—third-party liability for the carrier’s own stack.
Open NAIC AI topicGartner: Unmonitored AI scales faster than accountability
Revenue tied to unmonitored AI is projected to rise from about $150 billion to $1 trillion by 2028. That boom outpaces model-accountability programs—and shows up as behavioural drift in claims and underwriting automation.
Read Gartner coverageNAIC: Live AI Risk Evaluation Supplement pilot
Twelve participating states are piloting the tool through September 2026 so regulators can see how insurers use AI and where governance, training, or model risk may fail.
Read NAIC pilot summaryBuilt for insurance compliance desks
Pilot terms available upon request.
Every pilot is designed around insurer accountability—not vendor absolution. Detection occurs before the claims decision stack. Adverse action requires documented human review. Flagged files are exfiltrated from the internal AI stack and sandboxed so fraudulent interpolation cannot drift production models.
- NAIC Model Bulletin alignment Written AI Systems program: governance, vendor diligence, monitoring of adverse decisions by model, and documented human review.
- Verified automation Flags route to adjuster/SIU; adverse action requires documented substantive review.
- Audit-ready evidence Model version, input hash, feature rationale, and reviewer notes retained for exams.
- Data residency options Pilot data stays in agreed U.S. regions; retention mirrors carrier policy and state rules.
- Workflow-agnostic independence ARP layers into existing stacks without becoming the claims system of record.
- Model drift audit Adjuster and SIU review, AI system recall, and data sanitization when fake media has already entered training or automation.
Advisory governance
Insurance, fraud, and AI-stack integrity around the pilot.
Twenty-five U.S. jurisdictions have adopted the NAIC Model Bulletin; a 12-state AI Risk Evaluation Supplement pilot runs March–September 2026. ARP constitutes independent advisory review so carriers can show examiners how intake, vendor models, and drift are governed—without replacing the carrier’s own legal, compliance, or procurement sign-off.
Insurance compliance
Regulatory governance, vendor diligence, consumer impact, and examination readiness.
SIU & claims operations
Fraud workflows, adjuster escalation, evidence handling, and practical false-positive review.
AI stack & infrastructure
Model integrity, data controls, evidence stack, and sanitization after contaminated intake.
This is a compliance-incorporated advisory layer, not ARP’s statutory board of directors. It supplements—but does not replace—the carrier’s legal, compliance, security, and procurement review.
NAIC implementation map · April 1, 2026
Model bulletin adoption, state by state.
Click a legend item to isolate a category. Source: NAIC map for Use of Artificial Intelligence Systems by Insurers.
International signal
Catastrophe losses and the protection gap, in numbers.
Primary sources on insured vs uninsured disaster cost—treaty desks, cedents, and emerging-market exposure.
Swiss Re sigma: $220B catastrophe losses, $107B insured
Natural-catastrophe economic losses were USD 220 billion in 2025; about 49% were insured (USD 107 billion)—the highest insured share on sigma records. In emerging economies, 80–90% of catastrophe losses typically go uninsured.
Read Swiss Re InstituteWorld Bank: Pakistan 2022 floods — $30B+ damages and losses
The PDNA estimates USD 14.9 billion in damages and USD 15.2 billion in economic losses (over USD 30 billion combined), with at least USD 16.3 billion needed for resilient reconstruction. 33 million people were affected.
Read World Bank assessmentCRED EM-DAT: 393 disasters, $242B economic damage
EM-DAT recorded 393 natural-hazard disasters in 2024: 16,753 deaths, 167.2 million people affected, and US$241.95 billion in economic losses.
Open CRED EM-DAT reportPrediction · human review · verification
Score the next event. A person still decides.
Mapped disaster history has informed insurance pricing since the 1930s. ARP modernizes that practice for reinsurance desks: geospatial and catastrophe records become risk scores a human can actually use—then someone reviews before any action.
Prediction
Past maps, future scores. Flood, fire, and quake records—plus current geospatial feeds—generate a forward risk score for the treaty, the cedent, and the region. The method is old. The data is current.
Human in the loop
AI assists. It does not bind, decline, or pay. ARP surfaces a score so underwriters and SIU can see the risk quickly; a person always reviews before anything moves. Built to be used—not decoded by a data-science team.
Audited information for data integrity
When a claim file arrives, check it. Controlled image capture plus cross-source corroboration—satellite stills, news, NASA and other open data—so the picture can be verified, or the claim analytically supported, before it contaminates a portfolio.
Initial pilot · accountable delivery
Access tracked. Model monitored. Drift audited.
Carrier teams work directly with ARP R&D. Intake is measured, loss is studied, and contaminated data is sandboxed—so automation stays verified rather than extrapolated from fraudulent files.
-
01
Access tracked
Where claim media enters FNOL. NAIC-bulletin compliant, AI-system tailored, state regulated.
-
02
Monitors model
Matrix score of provenance, artificial-data forensics, and C2PA + ARP Protocol.
-
03
Data trail logged
Cryptographic chain-of-custody, escalated risk tiers, and a defensible fraud packet.
-
04
Model drift audit
Adjuster and SIU review; AI system recalled; training and automation data sanitized.
“Policy decisions cannot rely on scattered solutions that compromise data integrity. ARP exists to give informed audits in unregulated data. Insurance leaders have the opportunity to control their future in world models for the AI revolution.”
Pilot interest intake
Request a compliance-ready pilot brief.
Submissions land in ARP’s secure intake database. ARP leadership reviews them—and our R&D leader will reach out to schedule a discovery call. Diligence packet available upon request.
- No auto-denial in the pilot charter—complete override authority
- Mutual NDA before discovery; call available within days of submitted interest
- NY / NJ / CT carriers prioritized for FY 2026 Q3 pilot cohort