Secure Multiparty Computation (MPC)

Takeaway

MPC lets parties compute a function on their private inputs without revealing them, learning only the output.

The problem (before → after)

Before: Joint analytics require pooling raw data, violating privacy and compliance.
After: Parties contribute encrypted/shares; protocols compute outputs securely under defined threat models (semi-honest or malicious).

Mental model first

Imagine each person holding a piece of a puzzle; only when all pieces are combined do you see the picture, yet no one ever sees others’ pieces alone.

Just-in-time concepts

Secret sharing (Shamir), garbled circuits (Yao), OT, additive shares.
Security models: semi-honest vs malicious adversaries.
Preprocessing: Beaver triples accelerate arithmetic circuits.

First-pass solution

Choose representation (Boolean vs arithmetic); compile function into gates; run protocols (GC/OT or secret-sharing based) with preprocessing for efficiency.

Hybrid with HE and TEE for best-of-all-worlds.
Circuit optimizations: depth, communication rounds, and batching.
Differential privacy on outputs to mitigate leakage.

Principles, not prescriptions

Match protocol to function structure and network conditions.
Define threat model and trust assumptions upfront.

Common pitfalls

Side-channel leaks from timing or abort patterns.
Incorrect output handling leaks too much information.

Connections and contrasts

See also: [/blog/homomorphic-encryption], [/blog/zero-knowledge-proofs], [/blog/differential-privacy].

Quick checks

Why preprocessing? — Moves heavy crypto offline to speed up online phase.
When garbled circuits? — Boolean logic-heavy tasks or two-party settings.
When secret sharing? — Arithmetic-heavy tasks with many parties.