ShieldAi

README (3).md

🛡️ ShieldAI — AI Agent Security Layer

$45M+ lost to AI agent exploits in Q1 2026. ShieldAI is the fix.

License Network Status Layers SDK

ShieldAI is the first multi-layer runtime security platform for autonomous AI agents operating in crypto and DeFi environments.

Unlike code auditors (pre-deployment) or wallets (custody), ShieldAI enforces security at runtime — while the agent is executing.

🚨 The Problem

In Q1 2026, over $45M was lost to AI agent exploits:

Traditional security tools were built for deterministic code. They don’t understand semantic intent, can’t detect prompt injection, and have no concept of behavioral drift in an AI agent context.

ShieldAI was built specifically for this gap.

✅ The Solution — 13 Security Layers

Layer Name Function Threat Covered
L0 MCP Gateway Intercepts all MCP tool calls before execution Data exfiltration, unknown servers, tool poisoning
L1 Input Firewall Scans all external data entering the agent Prompt injection, hidden text, unicode spoofing
L2 Memory Protection Cryptographic hash chain for agent memory Memory poisoning, rule injection, history tampering
L3 Intent Checker Claude API semantic analysis of agent actions Scope creep, behavioral inconsistency, social engineering
L4 TX Guardian Pre-chain transaction simulation and validation Wallet drainers, honeypots, scam addresses
L5 Agent Watchdog Behavioral baseline + anomaly detection Behavioral drift, burst attacks, off-hours activity
L6 Human Gate Hardware-attested biometric authorization High-value transactions ($10K+), critical operations
L7 Rate Limiter Per-agent transaction rate enforcement Flood attacks, rapid drain attempts
L8 Whitelist Engine Dynamic counterparty whitelisting Unauthorized recipients, new addresses
L9 Freeze Circuit Emergency agent freeze mechanism Active exploits, anomalous behavior patterns
L10 SpendGuard On-chain spend limit enforcement (Solidity) Overspend, daily limit breaches
L11 EU AI Act Logger Compliance logging for Articles 9,13,14,15,17,72 Regulatory non-compliance, audit failures
L12 Audit Trail Immutable on-chain event log Tamper attempts, accountability gaps

🎮 Interactive Demos

Every layer has a live browser demo — no installation required:

Layer Demo Description
L0 MCP Gateway → Demo Intercept and inspect MCP tool calls in real-time
L1 Input Firewall → Demo Test 13 prompt injection detection patterns
L2 Memory Protection → Demo Simulate memory poisoning attacks and defenses
L3 Intent Checker → Demo AI-powered semantic intent analysis (live Claude API)
L4 TX Guardian → Demo Simulate transaction validation and drainer detection
L5 Agent Watchdog → Demo Behavioral anomaly scoring and auto-freeze
L6 Human Gate → Demo Mobile biometric authorization flow
L7 Rate Limiter → Demo Real-time rate limiting simulation
L8 Whitelist Engine → Demo Dynamic whitelist management
L9 Freeze Circuit → Demo Emergency freeze and recovery flow
L10 SpendGuard → Demo On-chain spend limit enforcement
L11 EU AI Act → Demo Compliance report generator
L12 Audit Trail → Demo Immutable on-chain audit log

→ Main Dashboard

🚀 Deployments

Contract Network Address Explorer
SpendGuard.sol Mantle Sepolia Testnet 0x7f875B92c772C48281a901C4BF32b53d6329fadC View ↗
SpendGuard.sol Arbitrum One (Mainnet) 0xE03C389DF391549E44c2aa807576c9eE2956C2d8 View ↗

📦 SDK

npm install @shieldai-xyz/sdk

import { ShieldAI } from '@shieldai-xyz/sdk';

const shield = new ShieldAI({
  contractAddress: '0x7f875B92c772C48281a901C4BF32b53d6329fadC',
  network: 'mantle-sepolia'
});

const result = await shield.checkSpend({
  agent: agentAddress,
  amount: transactionAmount,
  token: 'USDC'
});

if (!result.allowed) {
  console.log('Blocked:', result.reason);
}

⚖️ EU AI Act Compliance

ShieldAI’s L11 EU AI Act Logger is our primary differentiator — no direct competitors exist in this space.

Covers Articles: 9 · 13 · 14 · 15 · 17 · 72

🏆 Competitive Positioning

  ShieldAI Claw Wallet Octane
Runtime monitoring ✅ ❌ ❌
Spend enforcement on-chain ✅ ⚠️ custody only ❌
EU AI Act compliance ✅ ❌ ❌
13-layer architecture ✅ ❌ ❌
Code auditing ❌ ❌ ✅

🌐 Links

📄 License

MIT — Built by ShieldAI Lab