Skip to content
QORECHAIN
Register

Post-Quantum AI Agents

QoreGuardian

The post-quantum bodyguard for your funds.

Share on X

About

QoreGuardian is an autonomous post-quantum treasury guardian built on QoreChain. Unlike AI agents designed to spend under soft limits, QoreGuardian acts as an always-on on-chain bodyguard that detects threats and locks down funds before prompt injection, model misbehavior, or quantum key compromise ("Q-Day") can drain a wallet. Its 4-layer defense-in-depth architecture combines: (1) QoreChain Scoped EVM Windows restricting agent execution strictly to GuardianAccount; (2) on-chain per-tx, daily, and allowlist caps with automated limit reduction, account freezing, force-evacuation, and an owner-triggered Dead-Man's Switch; (3) an explainable AI Risk Engine streaming live 0–100 threat scores and plain-English reasoning; and (4) a RecoveryVault that unlocks evacuated funds only with a fresh ML-DSA-87 post-quantum signature verified on-chain via QoreChain's pqcVerify precompile (0x...0A01).

How it's made

Approach & Core Idea Most Web3 AI agents are built to spend under soft guardrails. QoreGuardian flips that model: it is an autonomous, always-on treasury bodyguard designed to detect threats and lock down funds before prompt injection, model misbehavior, or a post-quantum key compromise ("Q-Day") can drain an account. 4-Layer Defense-in-Depth Architecture 1. Protocol Layer (QoreChain Scoped EVM Windows): Execution for the agent is constrained at the network level using openEvmWindowMsg with allowedTargets locked strictly to GuardianAccount, plus tight block, transaction, and fee caps. 2. Smart Account Layer (GuardianAccount.sol): Enforces hard on-chain per-transaction, daily, and allowlist spending limits. The agent can only tighten security (reduceLimits, freeze, forceMoveToVault) it can never raise its own limits. An owner-only emergencyEvacuate() Dead-Man’s Switch lets the human owner evacuate funds directly even if the AI goes offline. 3. AI Brain Layer (agent/): A Node.js + TypeScript service (monitor.ts, riskEngine.ts, defender.ts, server.ts) that watches live account activity, computes an explainable 0–100 Risk Score with plain-English reasoning, and triggers automated defensive escalation. 4. Post-Quantum Recovery Layer (RecoveryVault.sol): An emergency destination that only unlocks evacuated funds when the human owner provides a fresh ML-DSA-87 post-quantum signature verified on-chain via QoreChain’s native pqcVerify precompile (0x0000000000000000000000000000000000000A01). Kits & Network Features Used 1. Smart Contracts: Solidity + Foundry (GuardianAccount.sol, RecoveryVault.sol, IPQCVerify.sol), deployed to QoreChain Testnet (qorechain-diana). 2. Agent & SDKs: Node.js, TypeScript, @qorechain/sdk, and viem. 3. Frontend: Vite + React + TypeScript + @qorechain/sdk for the live monitoring dashboard and Q-Day / prompt-injection attack simulations. 4. QoreChain Native Features: Scoped EVM Windows (openEvmWindowMsg) and the pqcVerify precompile (0x...0A01) for on-chain ML-DSA-87 verification. What Was Hard 1. Asymmetric Privilege Design: Ensuring the agent has enough authority to freeze and evacuate funds during an active attack, while making it mathematically impossible for a compromised agent to raise spending caps or redirect RecoveryVault ownership. 2. Bridging Classical EVM & Post-Quantum Cryptography: Coordinating standard EVM account execution inside QoreChain’s scoped EVM windows with ML-DSA-87 signature verification through the pqcVerify precompile, including testing against a mock precompile in Foundry before live testnet verification. 4. Fast, Explainable Risk Scoring: Tuning the risk engine so real drain attempts and prompt-injection anomalies escalate cleanly (reduce limits → freeze → force-move) with clear human-readable explanations, without triggering false lockouts on normal transactions.

Timeline

QoreGuardianPost-Quantum AI Agents

New project.

Comments and questions

  • No comments yet.