Finance Deployments
Deployment evidence
Financial PET projects are often described through vendor case studies, consortium announcements, or research demonstrations. Treat maturity labels conservatively.
Vendor Case Studies And Unclear Production Claims
TriBank / Amlytic privacy-preserving financial-crime analytics
| Field | Entry |
|---|---|
| Organization / project | Amlytic, Secretarium, FutureFlow, and TriBank Initiative references |
| Domain | Finance |
| Problem | Pool cross-institution transaction intelligence for financial-crime detection without exposing raw institutional transaction data. |
| PETs used | Secure multiparty pseudonymisation, privacy-preserving matching, de-identified graph analytics |
| Deployment maturity | Vendor case study / unclear production maturity |
| Source quality | Vendor case study |
| What worked | Public materials describe use of privacy-preserving matching and analytics to combine financial-institution contributions. |
| Challenges | Source material is vendor-operated and does not provide an independent production evaluation, false-positive analysis, or detailed threat model. |
| Lessons for builders | Cross-bank analytics needs privacy-preserving entity resolution, governance over outputs, and careful distinction between de-identified analytics and formal privacy guarantees. |
| Source | Amlytic product and TriBank description |
(Evidence: Deployment-backed only as vendor-described. Source quality: Vendor case study. Reviewed 2026-06-17 — useful lead, but not independent evidence of production maturity or detection lift.)
Published Pilots And Demonstrations
Japanese banks federated fraud-detection demonstration
| Field | Entry |
|---|---|
| Organization / project | Demonstration experiment with five Japanese banks |
| Domain | Finance |
| Problem | Test privacy-preserving federated learning for fraudulent financial transaction detection using real transaction data across banks. |
| PETs used | Federated learning |
| Deployment maturity | Demonstration experiment |
| Source quality | Peer-reviewed / academic |
| What worked | The paper reports a multi-bank demonstration using real transaction data and names participating banks. |
| Challenges | The authors describe it as a demonstration experiment; it should not be treated as evidence of production deployment. |
| Lessons for builders | Fraud FL needs real data, cross-bank governance, and operational design beyond model training. |
| Source | IPSJ Journal of Information Processing: Privacy-Preserving Federated Learning for Detecting Fraudulent Financial Transactions in Japanese Banks |
(Evidence: Literature-backed. Source quality: Peer-reviewed / academic. Reviewed 2026-06-17 — the paper reports five named banks and real transaction data; production adoption remains unestablished.)
Common Proposed Use Cases
| Use case | Candidate PETs | Why proposed | Caveats |
|---|---|---|---|
| Cross-bank fraud detection | FL, secure aggregation, MPC, DP | Fraudsters move across institutions | Poisoning, false positives, and latency can matter more than raw accuracy |
| AML graph analysis | MPC, PSI, privacy-preserving pseudonymisation, clean rooms | Institutions need joint network signals | Entity resolution and output governance are hard |
| Credit-risk benchmarking | Federated analytics, DP, clean rooms | Banks want peer comparison without exposing portfolios | Aggregates can reveal business-sensitive information |
| Customer overlap analysis | PSI, clean rooms | Institutions need match counts or permitted overlaps | The intersection itself may be sensitive |
Lessons Learned
- Financial deployments need adversarial thinking: participants, customers, fraudsters, and insiders may all be strategic.
- Entity resolution is often the hidden hard problem.
- PETs can protect inputs while still producing outputs that reveal investigations or competitive data.
- Vendor case studies are useful leads, but production maturity and independent evaluation often remain unclear.