Deepfake protection for claim evidence

The defence against
deepfake fraud.

Generative AI can fake a claim photo or document in seconds - and it passes inspection. CapSeal is the protection against it. Evidence is sealed on the device the instant it's captured, so genuine claims pay instantly and deepfaked ones can never produce a valid seal. You stop fabricated evidence before it's paid, not after.

Click for sound · 13s
CLAIM #MTR-4471 / front-left
SEALED / AUTO-PAY
sha256:9f2a...c41e · SecureEnclave · CapSeal v1
98%
of insurers say AI editing tools are fuelling digital claim fraud
Verisk, State of Insurance Fraud 2026
only 32%
feel confident they can detect a deepfake in a claim
Verisk, 2026
20-30%
of claims may now include altered images or fabricated documents
Shift Technology estimate, 2026
The shift nobody's core system was built for

Detection came too late. So we moved upstream.

Every incumbent fraud tool asks the same question after the fact: does this uploaded photo look fake? Against generative AI, that's a losing arms race - fakes now pass traditional checks over 90% of the time. CapSeal asks a different question, and asks it at capture.

The old question post-hoc

"Here's a photo the claimant uploaded. Does it look real?"

Forensics guesses at pixels after the fact. Every new image model resets the game. You're always one release behind the fraudster, and the honest customer waits weeks while you check.

The CapSeal question at capture

"Can this photo prove it came from a real phone, at a real moment, unaltered since?"

That's not an opinion - it's math. A sealed photo is trustworthy by construction. Fabricated evidence can't produce the seal, so it never enters the fast lane in the first place.

Capture to decision, in one flow

Four steps. No new app for your customer.

CapSeal drops into the claims app you already run. Your customer photographs the damage the way they always have - the trust happens invisibly, on the device.

STEP 01

Capture

The claimant photographs the damage inside your existing app, powered by our capture SDK.

STEP 02

Seal

The image is signed on-device using the phone's hardware security chip - bound to device, time, and place before anything can touch it.

STEP 03

Verify

Our engine checks the seal and runs anti-spoofing: a real lens saw a real scene - not a screen, not an injected feed.

STEP 04

Decide

You get one verdict, with a proof you can re-check yourself.

AUTO-PAY · REVIEW · REJECT
Why you can trust the verdict

You don't trust us. You trust the math.

CapSeal removes itself from the trust equation. Every seal rests on three things your business already trusts without us - and your own engineers can verify each one independently.

01 / HARDWARE

Apple & Google's own chips

The "real, un-tampered device" guarantee comes from Secure Enclave and Play Integrity - built by Apple and Google, not by us.

02 / CRYPTOGRAPHY

The same math as your bank

Standard cryptographic signing. Change one pixel and the seal provably breaks. Your team can verify it on your own servers, offline.

03 / VERIFIABLE PROOF

Proof your auditors can check

Every verdict exports a cryptographic proof in a format your auditors and regulators can independently verify - while the engine that produces it stays exclusively ours.

None of this is asserted on trust. The adversary model, the protocols, the unforgeability bound and the limits we cannot design around are all set out in the technical whitepaper - including Theorem 3, which concedes what no capture system can do.

Why this is a business, not a feature

Proprietary technology. Verifiable proof.

CapSeal's sealing and anti-spoofing engine is proprietary - purpose-built for claims, adversarially hardened, and ours alone. Anyone can verify a CapSeal proof; only CapSeal can produce one. Four layers of technology no carrier, and no competitor, can simply lift.

1

Capture integrity

Defeat screen-of-screen, injected virtual cameras, and rooted devices. Our proprietary capture-integrity engine guarantees what the camera hands over is real - adversarial, insurance-specific, continuously hardened against attacks that haven't been published yet.

CORE IP
2

The verdict engine

Turn a raw seal into an auditable insurance decision - tuned to your lines, with graceful handling of no-chip phones, offline capture, and legacy uploads, so honest claims are never blocked.

3

The verification network

A sealed capture, verifiable across insurers, repairers and reinsurers - so one loss can't be sold twice. No single carrier can build this for its competitors.

DURABLE MOAT
4

The intelligence layer

Every verdict feeds an entity-graph engine that answers the question the seal can't: is the intent honest? Provenance for the evidence, intelligence for the intent.

Who it's for

One champion. One metric. One line of business.

THE BUYER

Head of Claims / Fraud

Signs on one number that moves on day one: the share of claims you can pay instantly without fraud risk. No consortium to assemble, no historical data to train.

THE INTEGRATOR

Your app team

Drops the SDK into the claims app you already ship. Days, not quarters, using primitives your security team already trusts.

THE END USER

The claimant

Photographs the damage the way they always have. The trust happens invisibly. Zero adoption friction, because there's nothing new to learn.

Start where fraud is loudest:
motor claims.

Highest photo volume, highest fraud, simplest integration. Prove it on one line, then expand across the book.

Book a pilot Motor brief (PDF) Technical whitepaper (PDF)
The pilot offer
Scope1 line of business (motor)
Duration60-90 days
IntegrationSDK drop-in, days not quarters
Success metricagreed up front, auto-pay rate
Pricingper claim verified + platform floor
How it works, in one picture

The mechanism behind the seal

Two things at the moment of capture - hardware attestation and a provenance seal - are what make the evidence impossible to deepfake.

STEP 1 Capture Photo taken in your app STEP 2 Seal Locked on the device STEP 3 Verify Seal + spoof checks STEP 4 Decide Pay / review / reject INSIDE THE SEAL Hardware attestation Apple and Google's own security chips prove a real, un-tampered phone and camera took the photo. Did a real camera see a real scene? + Provenance The photo is cryptographically bound to device, time and place. Change one pixel and it breaks. Can this photo prove its own origin? = Sealed evidence A photo that carries its own proof. Deepfakes can't produce it. Verifiable by anyone, without trusting us. Genuine claims pass and pay instantly. Fabricated evidence never gets a seal.