Secure Multi-Party Computation (SMPC)

The foundational property of SMPC is that no party learns anything about another party's private input beyond what can be logically inferred from the output of the computation itself. This is formally proven using the real-ideal paradigm, where the security of the real protocol is demonstrated by showing it is computationally indistinguishable from an ideal world where a perfectly trusted third party performs the calculation. In the context of financial fraud, this means Bank A can check if a suspicious account is on Bank B's blacklist without Bank B ever revealing its full list.

SMPC protocols ensure that the output of the distributed computation is mathematically identical to what would have been produced if a single, trusted entity had computed the function on all the plaintext inputs. This is critical for fraud detection integrity; a collaborative risk score must be as accurate as a centralized one. The protocol is designed to be robust against malicious adversaries who may attempt to deviate from the specification to corrupt the result, often using Message Authentication Codes (MACs) within the secret-sharing scheme to abort the computation if tampering is detected.

Most practical SMPC systems rely on linear secret-sharing schemes like Shamir's Secret Sharing or additive sharing. A private input value is split into random-looking shares distributed among computing parties. Individually, a share reveals no information about the original input. Computation is performed on these shares using distributed, interactive protocols:

• Addition: Parties can locally add their shares without communication.
• Multiplication: Requires one round of communication and pre-computed Beaver triples to mask intermediate values. This allows complex functions like neural network inference to be broken down into circuits of addition and multiplication gates.

An alternative paradigm to secret-sharing is Yao's Garbled Circuits, a constant-round protocol ideal for two-party computation. The function is represented as a Boolean circuit. The 'garbler' encrypts the truth tables of every logic gate, and the 'evaluator' uses Oblivious Transfer (OT) to retrieve the keys corresponding to their private input without revealing it. The evaluator then decrypts the circuit gate-by-gate to learn the final output. This approach is highly efficient for comparison-heavy fraud checks, such as threshold-based risk scoring, where the circuit depth is shallow.

SMPC protocols are architected for different threat models, which directly impacts performance:

• Honest-Majority (e.g., 3-party): Assumes a majority of computing parties follow the protocol. Enables extremely fast, information-theoretic security without heavy cryptography. Ideal for consortiums of trusted banks.
• Dishonest-Majority: Tolerates a scenario where only one party is honest. Requires computationally heavy, non-interactive zero-knowledge proofs to enforce correct behavior. This is necessary for open, permissionless fraud intelligence networks. The choice of model represents a direct trade-off between computational overhead and trust assumptions.

A critical security property for financial applications is fairness: if one party learns the output of the computation, all parties must learn it. Without fairness, a malicious party could abort the protocol after receiving the result, leaving honest parties without the collaborative fraud intelligence they need. Guaranteed Output Delivery (GOD) is a stronger property ensuring the computation completes regardless of adversarial behavior, often achieved through the use of a broadcast channel and redundant computation. This prevents denial-of-service attacks on the fraud detection pipeline.

A foundational two-party SMPC protocol where a function is represented as a Boolean circuit. The circuit generator encrypts (garbles) the truth tables of each gate. The evaluator receives keys corresponding to their input via Oblivious Transfer and evaluates the circuit gate-by-gate, learning only the final output. This approach is highly efficient for constant-round secure computation but incurs significant communication overhead for complex functions.

A method for distributing a secret value among multiple parties by splitting it into random shares. No single share reveals anything about the secret. The value can only be reconstructed when a sufficient threshold of shares is combined. In SMPC, additive secret sharing is commonly used, where shares sum to the original value. This enables linear operations (addition, scalar multiplication) to be performed locally on shares without any interaction between parties.

A fundamental two-party protocol where a sender inputs two messages, and a receiver selects and learns exactly one of them. The sender remains oblivious to which message was chosen, and the receiver learns nothing about the unselected message. OT is a complete primitive for secure computation, meaning any SMPC protocol can theoretically be constructed from it. Modern OT extension techniques dramatically reduce the number of expensive public-key operations required.

Pre-computed, random multiplicative triples used to enable efficient secure multiplication in secret-sharing-based SMPC protocols. A triple consists of random shares [a], [b], [c] where c = a * b. During the online phase, parties consume these triples to perform multiplications with only local computation and simple broadcast operations, avoiding expensive cryptographic interactions. Generating these triples offline is a key optimization for high-throughput fraud detection systems.

A specialized SMPC protocol enabling two or more parties to compute the intersection of their private datasets without revealing any elements outside that intersection. In financial fraud, PSI allows banks to discover shared money mule accounts or common suspicious entities across institutions without exposing their full customer lists. Variants include cardinality-only PSI (revealing only the intersection size) and threshold PSI (revealing the intersection only if it exceeds a minimum size).

A cryptographic scheme where a private key is split into shares distributed among multiple parties. A threshold t of parties must collaborate to perform a cryptographic operation (decryption or signing), preventing any single party or small coalition from abusing the key. In SMPC-based fraud analytics, threshold decryption ensures that aggregated risk scores can only be revealed when a quorum of participating institutions agrees, enforcing distributed trust and preventing unilateral data exposure.