Griffin

An investigation you can defend.

A Chief of Investigations. Griffin assigns the right investigator, conducts a disclosed interview by chat, voice, or video avatar, and hands your team a decision-ready file — every conclusion traced to an exhibit on an immutable record.

Built for the CSO, the MLRO, and the Head of Financial Crime

Case File / Exhibit Chain ● Immutable Record
CONCLUSIONS — TOUCH ONE C-01 · IDENTITY C-02 · SOURCE OF FUNDS C-03 · ADVERSE MEDIA EXHIBITS E-01 E-02 E-03 E-04 E-05 E-06 E-07 E-08 E-09 RECORD ⌗ 4A7F·D210·88EC·9C21 · APPEND-ONLY · NOTHING REWRITTEN
Representative case file — touch a conclusion to trace its evidence Every conclusion → an exhibit
By design
3Interview modes · chat, voice, video avatar
KYC·EDD·FRAUDCase types covered
100%Conclusions traced to exhibits
Append-onlyCase record, immutable by design
Griffin / Doctrine

Most investigations fail at the file, not the finding. A conclusion nobody can trace is a conclusion nobody can defend.

Disclosed — the subject knows they are speaking with an AI investigator. Recorded — every exchange lands on an append-only record. Traced — every conclusion points at its exhibit.

Griffin recommends. A named human decides, and signs. Disclosed · Auditable · Human-signed.

Three moves. One defensible file.

01

Assigns the investigator

The profile suited to the subject — chosen to build rapport, not to trick. Rapport produces better evidence.

02

Conducts the disclosed interview

Chat, voice, or video avatar — the subject’s choice, on their schedule. Every exchange becomes an exhibit.

03

Hands over the file

Decision-ready, every conclusion traced. Your team reviews, decides, signs. The audit trail is already written.

The only interview in the category with a scorecard.

Fifteen sealed scenarios with locked answer keys, across two disciplines — KYC/EDD and insider-risk security debriefing. Clean cases must clear: over-escalation counts as failure.

The benchmark has teeth. It has caught real design gaps and forced the fix before release — finding, fix, and proof kept on the record. A model-validation pack travels with every release.

FindingFixProofon the record, before release
Benchmark / Scorecard ● Every Release
Escalation discrimination Measured every release Escalate what should escalate — and clear what should clear.
Risk-band accuracy Measured every release Is the number in the right neighbourhood.
Finding recall Measured every release Did the memo surface what a human needs.
Fact-conflict recall Measured every release Did it catch answers that contradict ground truth.
Guardrail breaches 0 — every run Prohibited techniques are enforced in code, not tone guidance.
Interview craft Scored turn by turn Pressure subtracts. Current numbers shared in a briefing.
Sealed keys · never visible to the model · current numbers in a briefing

One engine, climbing a ladder of disciplines.

Interviewing is a craft that generalises. Six disciplines of human intelligence work, each assuming everything beneath it — and one supervisory layer spanning the stack.

The method: train to the apex, apply at the base. Skills built for the hardest rung make every rung below better.

Investigative Interviewing — Seeks and verifies facts. Builds rapport, files defensible reports.

Griffin supervises & orchestrates across the stack

The people who sign the decision.

CSO

A defensible record of every investigative decision — built into the process, not reconstructed after the fact.

MLRO

KYC and EDD interviews that scale without losing the audit trail the regulator will ask for.

Head of Financial Crime

Fraud interviews conducted consistently, on the subject’s schedule, with every exchange preserved as evidence.

Levenhall / Griffin / Briefings ● Open

See a disclosed investigation run end to end.