BenchmarksStack Ranking
APIsPricingStandardDocsWhite PaperTokenBlogAboutSecurity Demo
Log InTalk to UsGet API Key
Post-Quantum Encrypted Fraud Intelligence

Your fraud team sees your data.
Fraudsters see all of it.
Now your team can too.

FraudShield is the first cross-institutional fraud intelligence network built on fully homomorphic encryption. Competing banks and insurers share fraud signals without sharing a single byte of customer data.

Request Early Access See How It Works
<500ms
P99 threshold query latency
18%+
Incremental fraud detection lift
0
Bytes of plaintext data shared
13
API endpoints, one integration
The Problem

Cross-institutional fraud is the largest unsolved problem in financial services

Organized fraud rings operate across five banks simultaneously. No individual institution can see the coordinated pattern because fraud detection systems are siloed by design.

🔒

Privacy Law Blocks Sharing

GLBA, CCPA, and GDPR prohibit sharing raw transaction data across competing institutions. General counsels reject every proposal. FraudShield computes on encrypted ciphertext, so no plaintext data is shared in any legally meaningful sense.

🏦

Competitors Won't Cooperate

Banks are direct competitors who will never gift each other customer relationship data. FraudShield's cryptographic isolation means no member can extract another member's data through any query pattern.

💣

Synthetic Identity Rings

One stolen identity opens accounts at five banks simultaneously. No single bank sees the pattern. FraudShield's encrypted velocity and ring detection catch multi-institution attacks in real time.

How It Works

Computation on ciphertext. No plaintext ever moves.

FraudShield uses H33's BFV fully homomorphic encryption pipeline. Entity feature vectors are encrypted client-side in the member's HSM. The H33 computation layer performs similarity matching, ring detection, and velocity analysis entirely on encrypted data.

Step 1
Encrypt & Enroll
Member encrypts entity feature vectors (up to 512 dimensions) using BFV homomorphic encryption. Private key never leaves the member's HSM.
<50ms per entity
Step 2
Consortium Query
During payment authorization, the member submits an encrypted query. H33 computes similarity against the entire consortium pool on ciphertext.
P99 <500ms
Step 3
Encrypted Result
H33 returns an encrypted risk score (0–1000) with a ZK-STARK proof of correct computation. Only the querying member can decrypt the result.
ZK-STARK verified
Step 4
Member Decides
The member decrypts locally and applies their own decisioning rules. H33 provides intelligence. Members retain full autonomy. Circuit-breaker ensures zero service impact.
Full decisioning control
API Reference

13 endpoints. One integration. Five engineering days.

SDKs for Python, Java, Node.js, Go, and .NET. OAuth 2.0 authentication with HMAC-SHA256 API keys. All responses include Dilithium post-quantum signatures.

POST
/v1/fraud/entity/encrypt

Entity Encryption

Encrypt feature vectors (up to 512 dims). Batch up to 10,000 entities per call.

<50ms per entity
POST
/v1/fraud/entity/enroll

Entity Enrollment

Register encrypted entities into your consortium namespace. GDPR tombstone support.

DELETE
/v1/fraud/entity/{id}

Cryptographic Deletion

Destroy BFV key material. Returns ZK proof of erasure for CCPA/GDPR compliance.

<24h deadline
POST
/v1/fraud/query/threshold

Real-Time Threshold Query

The hot path. Called inline during payment authorization. Returns encrypted boolean + risk score.

P99 <500ms — hard SLA
POST
/v1/fraud/query/similarity

Full Similarity Scan

Top-N encrypted similarity scores ranked against the consortium pool. For investigations.

P99 <2,000ms
POST
/v1/fraud/query/ring-detection

Fraud Ring Detection

Submit up to 50 encrypted entities. Returns encrypted graph connectivity score across institutions.

P95 <5,000ms
POST
/v1/fraud/query/velocity

Cross-Institution Velocity

How many times has this entity appeared across consortium members in the past N days?

GET
/v1/fraud/intelligence/risk-score/{id}

H33 Risk Index

Aggregated 0–1000 risk score with confidence interval and signal breakdown.

POST
/v1/fraud/intelligence/alert

Confirmed Fraud Alert

Submit confirmed fraud events. Flags the entity as Known Bad Actor. Feeds the consortium flywheel.

GET
/v1/fraud/intelligence/consortium-health

Consortium Health

Privacy-preserving aggregate statistics. No member-specific data disclosed.

GET
/v1/fraud/audit/proof/{id}

ZK-STARK Computation Proof

Independently verifiable proof of correct computation. First-class compliance deliverable.

GET
/v1/fraud/audit/log

Immutable Audit Log

Chain-hashed, tamper-evident log. Paginated, date-range filtered. Exportable as signed PDF.

POST
/v1/fraud/compliance/deletion-request

Regulatory Deletion

GLBA, CCPA, GDPR cryptographic erasure. Returns signed deletion certificate with ZK proof.

<24h fulfillment

Built on the same FHE pipeline that processes 2.17 million authentications per second.

38.5µs
Per-entity latency
99.99%
Uptime SLA
11 9s
Data durability
Pricing

Three tiers. One network effect.

Every confirmed fraud alert makes the network smarter for everyone. Members who contribute more receive higher query quotas and lower per-query pricing.

Observer
$48K
per year
Query-only access. Validate the product before committing to full enrollment.
  • 50,000 queries per month
  • Real-time threshold queries
  • Full similarity scans
  • Ring detection & velocity
  • ZK-STARK computation proofs
  • Immutable audit log
  • Entity enrollment
  • Consortium governance
Start Pilot
$0.008 per query over 50K/month
Anchor Member
$480K
per year
Contributor access plus governance voting rights and co-marketing.
  • Unlimited queries
  • All Contributor features
  • Governance voting rights
  • Approve new members
  • Set detection thresholds
  • Dedicated customer success
  • All 5 language SDKs
  • Co-marketing rights
Contact Sales
$0.003 per query over 1M/month
Compliance

Built for the regulatory landscape your general counsel lives in

Every compliance requirement is a first-class API deliverable, not a footnote. The GLBA Opinion Letter ships before any technical integration begins.

GLBA

FHE architecture provides the legal basis for consortium data sharing. Outside counsel GLBA Compliance Opinion Letter included for every member.

CCPA / CPRA

Cryptographic erasure within 24 hours. Deletion certificate with ZK proof of erasure serves as the regulatory fulfillment record.

GDPR

EU member data processed only on EU-region FHE nodes. Full data subject rights (access, deletion, portability) via API.

SOC 2 Type II

Audit observation period initiated at General Availability. Report available for vendor due diligence at all member institutions.

PCI DSS

FHE provides the technical control for Requirement 3 when vectors derive from cardholder data. Full scope documentation provided.

BSA / AML

Ring detection output feeds SAR filing workflows. H33 does not file SARs. Members retain all BSA obligations. Clearly documented.

FAQ

Frequently asked questions about cross-bank fraud detection

How can competing banks share fraud data without violating GLBA?

FraudShield uses fully homomorphic encryption (FHE) so that no plaintext customer data is ever shared between institutions. Each bank encrypts its entity feature vectors client-side using BFV homomorphic encryption, and H33's computation layer performs similarity matching and ring detection entirely on ciphertext. Because no plaintext data moves between institutions, there is no "disclosure" in the legal sense under GLBA, CCPA, or GDPR. An outside counsel GLBA Compliance Opinion Letter is included for every consortium member before any technical integration begins.

What is homomorphic encryption fraud detection?

Homomorphic encryption fraud detection means running fraud scoring algorithms directly on encrypted data without ever decrypting it. FraudShield uses BFV fully homomorphic encryption to compute similarity scores, velocity checks, and fraud ring connectivity graphs entirely on ciphertext. The computation produces encrypted results that only the querying institution can decrypt with its own private key. H33 never sees plaintext transaction data, customer identities, or fraud signals at any point in the pipeline. Every result includes a ZK-STARK proof of correct computation.

How does cross-institutional fraud ring detection work?

FraudShield's ring detection API accepts up to 50 encrypted entity vectors and computes graph connectivity scores across the entire consortium pool on ciphertext. The system identifies when the same synthetic identity or closely related entities appear across multiple financial institutions without any member revealing their customer data to another member or to H33. Results are returned as encrypted graph scores with ZK-STARK proofs. Only the querying institution can decrypt and interpret the results.

Can FraudShield detect synthetic identity fraud across multiple banks?

Yes. Synthetic identity fraud is FraudShield's primary use case. A single fabricated identity may open accounts at five banks simultaneously, and no individual bank can see the coordinated pattern because fraud detection systems are siloed by design. FraudShield's encrypted velocity checking and similarity matching detect when the same entity features appear across multiple consortium members in a short time window. The cross-institution velocity API reveals how many times an entity has appeared across member institutions in the past N days, all computed on ciphertext.

What is the latency for real-time fraud scoring with FHE?

FraudShield delivers sub-500ms P99 latency for real-time threshold queries called inline during payment authorization. This is the hard SLA for the hot path. The underlying FHE pipeline processes entities at 38.5 microseconds each, built on the same engine that handles 2.17 million authentications per second. Full similarity scans complete in under 2 seconds (P99), and ring detection queries complete in under 5 seconds (P95). A circuit-breaker architecture ensures zero impact to payment authorization flows if latency spikes.

How does encrypted velocity checking prevent application fraud?

Encrypted velocity checking counts how many times a specific entity's encrypted feature vector has appeared across consortium members within a configurable time window. This catches application fraud patterns like rapid-fire account openings across multiple banks using the same stolen or synthetic identity. The entire computation runs on ciphertext, so no member learns which other institutions were queried or what their specific data contains. Results include an encrypted count and a ZK-STARK proof of correct computation that can be independently verified.

What happens if H33's servers are breached? Is fraud data exposed?

No. A breach of H33's infrastructure exposes only BFV ciphertext, which is mathematically indistinguishable from random noise without each member institution's private key. Private keys never leave the member's HSM. H33 operates exclusively on encrypted data and cannot decrypt any consortium member's entity vectors, query results, or fraud signals. Even a complete root compromise of H33's entire server infrastructure reveals zero plaintext customer data. The encryption is lattice-based with NIST post-quantum security, resistant to both classical and future quantum attacks.

How does FraudShield compare to NICE Actimize and Featurespace?

NICE Actimize and Featurespace are excellent single-institution fraud detection platforms that analyze one bank's data in isolation. FraudShield solves the cross-institutional problem that neither can address: detecting organized fraud rings spanning multiple competing banks without sharing customer data. FraudShield is complementary to existing fraud systems. It adds an 18%+ incremental detection lift on top of existing fraud detection stacks by providing consortium-wide encrypted intelligence that was previously impossible due to GLBA, CCPA, and GDPR privacy regulations.

What is a consortium fraud network?

A consortium fraud network is a group of financial institutions that pool fraud intelligence to detect cross-institutional threats like organized fraud rings and synthetic identity attacks. Traditional consortium models require data sharing agreements that conflict with GLBA, CCPA, and GDPR. FraudShield is the first consortium fraud network built on fully homomorphic encryption, eliminating the data sharing barrier entirely. Members contribute encrypted entity vectors and receive encrypted intelligence. No plaintext data ever moves between institutions or through H33's infrastructure.

How does the confirmed fraud alert flywheel work?

When a consortium member confirms a fraud event through the Confirmed Fraud Alert API, that entity is flagged as a Known Bad Actor in the encrypted consortium pool. Future threshold queries from any member institution will reflect this signal in their encrypted risk scores. Every confirmed fraud alert makes the network smarter for every member. Members who contribute more confirmed fraud alerts receive higher query quotas and lower per-query pricing, creating a positive feedback loop that incentivizes participation and continuously improves detection accuracy across the entire consortium.

Ready to see the full picture?

Schedule a 30-minute technical deep dive. We'll show you the encrypted query pipeline, walk through the SDK integration, and answer your general counsel's questions.

Request Early Access Talk to Sales