
domain
Anti-Money Laundering
Screening and monitoring agents whose decisions replay
AML enforcement punishes process failures: alerts closed without documented rationale, screening steps skipped, decisions nobody can attribute. Those are exactly the failure modes receipts eliminate. When agents assist with KYC research, sanctions screening triage, or SAR preparation, every disposition mints a signed receipt on the hash-chained spine, and every action binds to a passport, so "who closed this alert and on what basis" has a cryptographic answer.
Refusals matter as much as actions in this domain. When policy blocks an agent, RCX fails closed and issues a signed RefusalReceipt with a reason code. A file full of signed refusals is evidence that controls operate, which is the artefact examiners actually look for and rarely find.
Human gates keep judgement where regulation requires it: escalations and filings route through approvals that cannot auto-approve past a timeout. And because the trail is replayable, a look-back exercise or an examiner's sample becomes a matter of replaying receipts rather than reconstructing months of analyst activity from case-management notes.
The result is an AML programme where automation increases the quality of the audit trail instead of thinning it.