Zero-Knowledge Proof (ZKP)

A succinct, non-interactive zero-knowledge proof that requires a trusted setup phase. zk-SNARKs enable verification of computation integrity with very small proof sizes and rapid verification times.

• Proof Size: Constant, typically a few hundred bytes
• Verification Time: Milliseconds, independent of computation complexity
• Trade-off: Requires a one-time trusted setup ceremony where toxic waste must be securely destroyed
• Use Case: Verifying a transaction complies with anti-money laundering rules without revealing the sender, receiver, or amount

A scalable, transparent zero-knowledge proof that relies on collision-resistant hash functions instead of elliptic curve cryptography. zk-STARKs eliminate the trusted setup requirement entirely.

• Transparency: No trusted setup needed, relying purely on public randomness
• Post-Quantum Security: Resistant to attacks from quantum computers due to hash-based construction
• Scalability: Proof size grows logarithmically with computation size, making them ideal for complex fraud detection pipelines
• Trade-off: Larger proof sizes than zk-SNARKs (tens to hundreds of kilobytes)

A cryptographic protocol enabling multiple banks to jointly compute fraud scores over their private transaction ledgers without revealing individual customer data to competitors.

• Input Privacy: Each party's raw transaction data remains encrypted and invisible to all other participants
• Collaborative Intelligence: Enables detection of cross-institutional fraud rings that no single bank could identify alone
• Protocol Overhead: Communication complexity scales with the number of parties, requiring optimized network topologies
• Real-World Application: Consortium of financial institutions computing shared risk scores for sanctioned entity detection

A cryptographic method that allows computations directly on encrypted data, generating an encrypted result that matches operations on plaintext. This enables fraud models to score transactions without ever decrypting them.

• Fully Homomorphic Encryption (FHE): Supports arbitrary computations on ciphertexts, though with significant computational overhead
• Partially Homomorphic Encryption (PHE): Supports only addition or multiplication, but is orders of magnitude faster
• Inference on Encrypted Data: A fraud detection model can run inference on an encrypted transaction and return an encrypted risk score
• Latency Consideration: FHE operations are 1,000x to 1,000,000x slower than plaintext, requiring hardware acceleration for real-time scoring

A mathematical framework for quantifying privacy guarantees in data analysis. When applied to fraud model training, it ensures that the presence or absence of any single transaction cannot be inferred from the model's outputs.

• Epsilon (ε) Budget: Controls the privacy-loss parameter; lower values (ε < 1) provide stronger privacy but reduce model accuracy
• Gaussian Mechanism: Adds calibrated noise to model gradients during training to mask individual contributions
• Composability: Privacy loss accumulates across multiple queries, requiring careful budget management in iterative training
• Regulatory Alignment: Provides provable compliance with GDPR and CCPA data minimization principles

A protocol allowing two financial institutions to discover common customers or flagged entities without revealing their full customer lists to each other.

• Use Case: Two banks identifying shared accounts linked to a fraud ring without exposing their entire customer base
• Efficiency: Modern PSI protocols based on oblivious transfer can intersect sets of millions of records in seconds
• Output Options: Can reveal only the intersection size, the intersecting elements, or associated metadata depending on the threat model
• Integration: Often combined with ZKPs to prove properties about the intersecting elements without revealing the elements themselves