The Last Bastion

Agent trust verification for the machine-to-machine economy. Agents prove who they are and what they carry — verified by cryptography, not promises.

The Problem

AI agents are deployed with API keys and system prompts. There's no standard way to verify an agent's identity across platforms, or check if the data it carries has been tampered with. NIST launched the AI Agent Standards Initiative in 2026 because this gap exists.

What It Does

Agent Trust Verification — 10-check pipeline:

• Ed25519 cryptographic challenge-response
• Behavioral analysis and anti-Sybil detection
• Signed passport (JWT) other systems can verify offline
• Blockchain-anchored proof on Polygon testnet

Payload Integrity Verification — 5-layer pipeline:

• Schema validation and injection detection
• Forensic image analysis (ELA, noise, copy-move)
• Adversarial challenge — actively tries to disprove results
• Tamper-evident Merkle-chain proof ledger

Stack

FastAPI · PostgreSQL · Redis · Ed25519 · Solidity · A2A Protocol · Python SDK · Docker

This is a working prototype built by a solo dev from NZ. Everything runs locally via Docker. Looking for feedback.