Evidence verification

Trust the evidence.
Before the decision.

We detect AI-generated and altered photos and documents, and show you where and why.

Starting with insurance claims.

See the demoHow it works
01A claim file comes in.
02Terrier scans it top to bottom.
03It zooms in where something doesn’t fit.
04It marks the exact spot and says why.
99%

of surveyed US claims professionals have encountered manipulated or AI-altered claim documentation

VERISK, STATE OF INSURANCE FRAUD, 2026
Editing a photo or a document now takes a minute and no skill. The checks most pipelines run were built for a world where it didn’t.

See it on a file

Three demonstration files, edited by us. Switch between the original, the edited version and what Terrier sees.

Sample file
Altered
REGION
Zoomed region
REASON
    CONFIDENCE
    Sample files edited by us. Highlighted regions and notes are illustrative until the prototype’s real output replaces them.

    How Terrier looks at a file

    Most detectors return a score. Terrier works like an investigator: it scans the whole file, stops where the picture doesn’t fit, looks closer, and writes down why.

    Policy document
    01 Scan the whole file
    →
    Attention map on the policy document
    02 Stop where it doesn’t fit
    →
    Zoomed date field
    03 Look closer
    →
    FINDING
    Altered
    WHERE
    Policy effective date
    WHY
    Field re-rendered in a different font; coverage now begins before the reported loss
    04 Explain
    Claimant submits→Insurer pre-screens→Terrier verifies→Claims decision

    One call inside the review flow you already run. The decision stays with the insurer.

    The same check applies wherever money moves on evidence: insurance claims today, returns and refunds next, identity decisions later.

    Founders

    Mathison

    Mathison

    Cofounder, technology

    Built the production systems at Alibaba that decide whether a face, an image or a document is real, at over ten million checks a day. Research on multimodal forgery detection. MSc in AI.

    LinkedIn
    Cassie

    Cassie

    Cofounder, insurance and go-to-market

    Actuarial science at Columbia; president of the Columbia Actuarial Society. Pricing, consulting and asset-liability work at three insurers. SOA candidate, six exams passed.

    LinkedIn

    Prove it on your own files

    We run on your closed cases, compare against what you do today, and agree the detection and review-load targets before any deployment.

    WEEKS 1–2

    Closed cases

    Fifty historical files with known outcomes. We report what we catch and what we miss, against your current process.

    WEEKS 3–6

    Shadow mode

    Live intake, no effect on payouts. We measure false positives on honest claimants and new leads found.

    THEN

    Deploy

    Per-file pricing or an annual agreement, with a false-positive budget you set. Integration is one call in your existing review flow.

    Start a pilot conversationGet a live demo