Summary
FinHarbor launched an AI co-investigator for AML — a self-hosted LLM that handles routine investigations while keeping humans in control of regulatory decisions. Designed for EU AI Act compliance from day one. Addresses the 95%+ false positive rate in rules-based AML systems.
Key Points
- Problem: 95%+ of AML alerts are false positives; 98% never result in SAR
- Solution: self-hosted LLM connected to ledger, KYC/KYB, KYT, transaction monitoring, audit trail
- Capabilities: plain language queries, false positive clearance, case assembly, SAR drafting, queue prioritization
- Human oversight: AI investigates, humans decide; no regulatory action without human signature
- EU AI Act compliance: documented, supervised, architecturally incapable of acting alone
- DORA compliance: append-only audit log meets third-party oversight; streams to client SIEM
- Data residency: deployed inside client's perimeter; regulated data never leaves
- Jurisdictions covered: GDPR, UK GDPR, revFADP, LGPD, PDPL
- Setup: scope access, select model, configure redaction rules
- Industry benchmarks: HSBC + Google Cloud cut alert volumes 60%+; Coforge reduced false positives 70%
- CEO Podoynitsyn: "Compliance teams don't need another dashboard — they need the routine taken off their hands"
- Broader trend: AI in compliance moving from experimental to production
- Key question: can AI reduce compliance costs without creating new regulatory risks?