ArbitrAI
Trustless dispute resolution with three AI judges, Confidential HTTP evidence, zero humans.
What it is
ArbitrAI solves the fundamental problem of peer-to-peer trust in Web3: when a client refuses to pay, there is nowhere to go. Legal fees exceed dispute value. Centralized platforms are opaque and biased. There is no trustless alternative.
ArbitrAI uses three independent AI models (Claude, GPT-4, Mistral) orchestrated by a Chainlink CRE workflow to resolve disputes automatically. Both parties lock ETH symmetrically in escrow where neither can walk away. Evidence is submitted as a content hash committed on-chain; the actual content is retrieved via Confidential HTTP inside a TEE, meaning no oracle node operator can read it. Three AI models independently evaluate the evidence and vote. A 2-of-3 consensus releases escrow to the winning party. If consensus cannot be reached, a circuit breaker refunds both parties.
The result is a fully autonomous arbitration system with no human moderator, no appeals process, and no possibility of bribery or bias dominating the outcome.
How it Works
- Smart contracts: Solidity (DisputeEscrow, ArbitrationRegistry, CREVerifier) on Sepolia
- CRE Workflow: TypeScript — Confidential HTTP evidence retrieval, parallel AI model queries, 2/3 consensus logic, ECDSA signature verification
- Evidence server: Node.js + Express for confidential evidence storage
- Frontend: React + Vite deployed on Vercel
- AI: Claude (Anthropic), GPT-4o (OpenAI), Mistral Large
- Stack: viem, Hardhat, Tailwind CSS
Links
Created by
- Max Karlsson