VeraLink
Privacy-preserving multi-institution AML and fraud co-detection.
Detect cross-institution fraud patterns without sharing raw records or PII. VeraLink gives regulated institutions a way to collaborate on encrypted-state analysis while preserving governance, traceability, and operational control.
Recall +10-20pp
Targeted improvement against cross-institution fraud patterns such as mule chains and synthetic identities.
p95 < 60s
Target end-to-end latency from institution submission to alert generation.
Consensus-gated disclosure
Only minimal summary metadata is revealed for threshold-exceeding cases.
Cross-institution crime exploits exactly what teams cannot safely share.
Fraud and AML risk increasingly live in the gaps between institutions. The challenge is not only identifying those patterns, but doing so without violating privacy, regulation, or data-sovereignty requirements.
Encrypted collaboration, minimal disclosure.
Each institution submits only behavioral and relational features encrypted with FHE. VeraLink performs statistics, similarity, and approximate inference directly on encrypted state, then reveals only minimal summary metadata for threshold-exceeding cases under governed decryption.
Regulation and commercial secrecy make raw-record or PII sharing impractical. Single-institution models remain blind to cross-institution fraud patterns.
Multi-hop transfers, mule networks, and synthetic identities span multiple institutions. The signal is distributed exactly where current teams cannot safely aggregate it.
On-prem systems stay siloed, PSI is weak for similarity and approximate detection, and onchain analytics miss critical offchain account and device signals.
A privacy system designed for regulated collaboration
VeraLink combines encrypted inference, threshold decryption, and audit commits into one operating model.
Encrypted inference
VeraLink runs linear and polynomial approximation ensembles, vector similarity, and cross-institution counts directly on encrypted feature sets.
Threshold decryption
Only summary metadata for threshold-exceeding cases can be revealed through 2-of-3 or 3-of-5 consensus. Raw data and per-institution contributions remain private.
Onchain audit commit
Policy, model, and alert hashes are committed to a permissioned chain, preserving an immutable audit trail without exposing operational data.
Target outcomes
Designed around measurable value for fraud operations, compliance, and governance.
Detection performance
Expected uplift for cross-institution fraud pattern detection.
Cleaner review queues and lower analyst overhead.
Operational performance
Target runtime for encrypted-state inference over shared feature batches.
Target from institution submission through alert delivery.
Security governance
No raw records or customer identifiers leave the institution.
Decryption is allowed only under threshold key agreement.
How VeraLink works
A four-step workflow from encrypted feature submission to governed alert disclosure.
Step 01
Local preprocessing and encryption
Each institution summarizes records into a standard feature schema and encrypts the features locally before submission.
Step 02
Encrypted-state computation
The central service computes aggregates, variance, similarity, and approximate inference without ever decrypting submitted data.
Step 03
Threshold-gated summary reveal
Only cases above the risk threshold can unlock minimal metadata such as reason code or reference key through threshold-key consensus.
Step 04
Asynchronous audit commit
Policy versions, model identifiers, and alert hashes are committed asynchronously for immutable auditability.
Market positioning
VeraLink sits between cryptographic primitives and real operational AML workflows.
Strong for blockchain traces, weak for offchain account, user, and device signals.
Still constrained by the inability to safely share cross-institution raw data.
Good at set intersections, weaker at similarity scoring and approximate detection.
Can support distributed compute, but often favors narrower or more linear workloads.
Powerful primitives, but no packaged AML product workflow for regulated teams.
Cross-institution approximate detection
Connectivity scoring across institutions
Threshold decryption by consensus
Permissioned onchain audit commits
Standardization strategy
VeraLink is not only a technical stack. It is also a path toward standard feature schemas, alert formats, and audit artifacts that regulators and institutions can trust.
Feature Schema 1.0
Publish a draft standard feature schema grounded in de-identification principles with early design partners.
Alert format and reason-code taxonomy
Define a standard alert envelope and reason-code structure for downstream review and easier information exchange.
Audit report schema
Create a regulator-ready audit report schema built around committed policy, version, and alert hashes.
Joint industry guidance
Work with associations and supervisory stakeholders on shared guidance for FHE-enabled AML and fraud prevention.
Roadmap
Define the standard feature schema
Build the base FHE operations pipeline
Design 2-of-3 threshold decryption
Ship the first onchain audit PoC
Add polynomial approximation ensemble models
Introduce cross-institution connectivity counting
Ship a policy DSL for governance
Provide an SDK for institution integrations
Optimize toward p95 < 5s encrypted inference
Define region-specific key governance policies
Automate audit report generation
Prepare for APAC and broader global rollout
Partners and R&D
Research-led development with design-partner input from regulated workflows.
VeraLink is being shaped alongside CryptoLab across FHE inference and bootstrapping tuning, threshold decryption, and key-governance design to resolve the hardest cryptographic bottlenecks early.
A draft feature schema, a PoC for encrypted sums, averages, and similarity, plus threshold decryption and onchain commit design are already in place. Initial outreach is focused on custody, prime, and neobank use cases.
Become a design partner
Bring your regulatory use case into the product definition and help shape the first production-ready feature schema and workflow model.
Frequently asked questions
Ready to scope a pilot?
Share your workflow, regulated regions, and data categories. We will frame a pilot scope, governance model, and integration path through the existing REDSAW contact flow.