Confidential computing

Compute together, share nothing

Confidential computing is the ability for several parties to jointly compute a function over their inputs while keeping each input private. Each party learns the result and nothing else. No single party (including Partisia) ever sees the raw data. Multiparty computation is the cryptography we use to do it.

In our work since the early 2000s

How it works

The intuition in 3 paragraphs

MPC splits every input into secret shares (unintelligible parts) and gives each share to a different node. Because no node holds more than its own share of any input, no node can reconstruct anyone's data: confidentiality comes from the math, not from trusting a party or a contract.

The nodes compute directly on the shares. Each node can add locally; multiplication and other operations take rounds of secure communication between nodes. Any function can be expressed this way (a sum, a comparison, a maximum, an equality check), and only the agreed result is revealed: to a chosen set of parties, to everyone, or to no one (kept secret-shared for a later computation).

Partisia builds on Shamir secret sharing with a configurable security threshold: the share of nodes an attacker would have to compromise before confidentiality is at risk. Stay below it and the original inputs are never reconstructed at any single point, even as the nodes exchange values to reach the result.

Where it shows up