Overview
Anti-Money Laundering (AML) screening verifies people and organizations against sanctions, PEP, watchlists, and adverse media to prevent the use of financial services for illicit purposes. Zyphe integrates AML screening with its decentralized Know Your Customer (KYC) solution to deliver end‑to‑end compliance from onboarding through ongoing monitoring. We screen identities across multiple lists and compute a risk score to help you make fast, defensible decisions.
Zyphe’s approach enables privacy‑preserving compliance. Instead of centrally storing sensitive personal data, Zyphe keeps user data decentralized and under the user’s control. Only the minimum necessary attributes or cryptographic credentials are shared with the screening engine and third‑party data sources (e.g., sanctions, PEP, adverse media providers). This achieves data minimization while maintaining high screening quality. Screening events and outcomes can be logged with strong cryptography and, where applicable, blockchain technology to provide an immutable audit trail for regulators and partners.
AML Agent Mode
When a screening produces PEP, sanctions, or adverse-media hits, results can be moderated manually or with help from the AML AI agent. Behind that agent, moderation is driven by a multi-role agentic review engine: analytic roles argue the case, a deterministic guard arbitrates auto-clear, and a tool-constrained actuator enacts one audited outcome. AML Agent Mode on the flow step that produced the screening controls how far that automation goes:
- Disabled: the agent is not invoked. Operators review every hit manually.
- Suggest: the agent runs and surfaces its verdict and rationale. Escalations are enacted so cases reach the human review queue; clears are suggestions only, a human confirms or rejects before the result is cleared.
- Auto-pilot: the agent may clear or escalate automatically. A clear is enacted only when the deterministic guard also passes; anything the guard vetoes is escalated for human review.
The mode is configured per step: on document verification for person screenings, and on the KYB step for company and UBO screenings. Missing or unparseable configuration is treated as Disabled (fail closed).
Auto-pilot reduces manual review effort by clearing low-risk false positives that pass the guard, while keeping high-confidence sanctions hits, Critical risk, thin identity data, and human verdicts under operator control.
For how the engine reaches and enacts each decision (the analytic roles, the deterministic guard, the audited action, and the safety guarantees), see AML Agentic Review Engine.
AML screening is one control in the wider antifraud map: document authenticity, liveness, KYB register checks, UBO unwrapping, device signals, and Transaction Monitoring sit alongside it. The same Disabled / Suggest / Auto-pilot modes are used on KYB and are intended for TM alert triage.