Every case in.
One decision out.
Ward is the decision model for fraud, AML, sanctions and KYC.
[01] We took the narrow road
AI that answers with a paragraph, a maybe a decision you can defend.
01 / decision
Typed, not written
Approve, step up or decline, in the same envelope for every case. Your code branches on one field.
02 / risks
A probability for every risk
Fraud, AML and sanctions each come back with their own probability and interval. A close call looks like a close call.
03 / evidence
Shown, not claimed
The signals, the lists checked and the typologies behind every answer, ready for your analyst and your regulator.
48 illustrative cases, one probability each.
[02] Confidence
Draw your own lines.
Every answer carries a probability and an interval, so your system decides what happens at each level: approve on its own, send the case to an analyst, or decline.
Move the two lines to see how the same cases split.
[03] See it decide
Pick a case.
Five of the eleven decision types. Same endpoint, same answer shape.
[04] Capabilities
Eleven questions. One endpoint.
Send what you have. Ward works out what it is, runs every check that applies and answers, in one call.
Should this payment go through?
Card, account-to-account or crypto. Started by a person, a merchant or an AI agent.
APPROVE · STEP_UP · DECLINE type: screenIs this name on a list?
Sanctions, PEP and adverse media, checked in one pass.
strong · potential · namesake · clear type: kycIs this applicant real?
Email, phone, document and lists, from your own onboarding flow.
VERIFIED · REVIEW · DECLINED type: documentIs this ID genuine?
Passports and IDs: check digits, alteration signals, holder screened.
consistent · review · altered type: statementWhat does this account history show?
Any file, any layout. Every transaction read, every counterparty screened.
STM_STRUCTURING · STM_PASS_THROUGH type: atoWas this account taken over?
The activity before a disputed transaction, read for signs of takeover.
new device · new payee · changed contact type: merchantCan we trust this merchant?
Website against declared category, owners screened, WMID.
website · owners · WMID type: creditHow does this credit case read?
The way a credit committee reads it, with factors and weights.
factors · weights type: intentIs this agent within its mandate?
Approve, step up or decline an AI agent's action before it pays.
mandate limits · EDQS-B type: agentCan this agent be trusted?
KYA Score and trust zone.
GREEN · AMBER · RED · CRITICAL type: askWhy did Ward say that?
Ask questions about any result, answered from its data.
see all capabilities →[06] Research
The science Ward grew out of.
It began with one question: how do you trust an AI agent with money?
| Paper | What it does | In Ward | Published |
|---|---|---|---|
| Economic Decision Quality Score | Grades an agent’s reasoning, not only its result. | EDQS-B is live EDQS-Q is in validation | Aug 2026 |
| Agent Behavioral Telemetry | Drift as an early sign of compromise. | Live | Jun 2026 |
| The Decision Trust Protocol | Seven decision gates and two system steps. | Live | May 2026 |
| The swarm that wasn’t | Laundering through one’s own agents. | Field note | Sep 2026 |