---
title: Chain-key cryptography
slug: chain-key-cryptography
description: The threshold-cryptography engine that gives each subnet its own signing key — without any single party ever holding the private material.
tags:
  - cryptography
  - core-concept
  - chain-fusion
date: 2026-03-21
updated: 2026-04-25
related:
  - internet-computer-protocol
hub: true
---

Chain-key cryptography is the engine that lets the Internet Computer sign
for itself. Each subnet has a stable public key; the corresponding private
key is split into shares held by the nodes, and never reconstructed in one
place. Signatures are produced by a threshold protocol — most nodes agree,
the protocol emits a signature, and from outside it looks like a single
trusted party.

This is what enables three otherwise-impossible things:

1. **Self-signing canisters** &mdash; a canister can ask the protocol to
   sign on its behalf. The result is a real signature that any external
   verifier accepts, derived from a key that no human ever held.
2. **Native multi-chain interaction** (Chain Fusion) &mdash; canisters can
   sign Bitcoin, Ethereum, and Solana transactions directly, without a
   bridge. Custody happens inside the protocol.
3. **Stable subnet identity** &mdash; subnets can rotate their underlying
   nodes without changing their public key, so an app&rsquo;s external
   identity stays stable even as the hardware beneath rotates.

## The protocol layer

The threshold scheme used is a derivative of BLS (Boneh-Lynn-Shacham)
signatures, with modifications for the network&rsquo;s specific needs:
asynchronous consensus, dynamic subnet membership, and the ability to
re-share secrets when nodes join or leave a subnet.

The cryptographic story is documented in the
[`Document library`](/document-library/#documents?tag=Technical%20Paper) under
Technical Papers. Look for entries on **DKG** (distributed key generation),
**threshold ECDSA**, and **vetKeys**.

## Why it matters for the editorial framing

The chain-key engine is what justifies calling the platform &ldquo;tamperproof.&rdquo;
There is no admin machine to compromise because there is no admin machine.
Even if an attacker physically takes a node, they hold a fragment of every
subnet key — useless on its own. The threshold protocol&rsquo;s safety
margin is configurable per subnet, but typically allows up to a third of
nodes to be Byzantine-faulty without compromising signing.

## Related

- [`Internet Computer Protocol`](/wiki/internet-computer-protocol/) &mdash; the network protocol the engine runs inside.
