ArxCode Whitepaper

Abstract

A purely private form of AI computation would allow builders to generate software, documents, and analysis without any intermediary observing, storing, or training on their inputs. Existing AI tools offer privacy as policy — a contractual promise enforced by the very party that profits from breaking it. We propose a protocol where privacy is enforced by architecture rather than by promise.

Work submitted to the network runs inside a sealed execution environment; inputs enter and finished output returns, but nothing in between is observable by the operator, the network, or any third party. Payment is settled per unit of compute using $ARX, an SPL token on Solana, eliminating accounts, subscriptions, and the data-for-service exchange that underlies conventional AI products.

Tokens are removed from circulation as work is performed and minted only when verified compute is delivered, binding supply directly to real usage. As long as honest node operators control a majority of compute capacity, the network produces correct output without exposing the data that produced it.