AgentGuard — AI Agent Transaction Fraud Prevention & Trust Layer for EU Merchants

The Problem

Agentic commerce — where AI agents autonomously browse, compare, negotiate, and purchase on behalf of consumers — is exploding. Visa launched "Intelligent Commerce," Mastercard launched "Agent Pay," Stripe released the Agentic Commerce Protocol. McKinsey projects $1T in agent-orchestrated revenue by 2030.

But no one is protecting merchants from the new fraud vectors this creates:

1. Agent Impersonation — Malicious actors create fake "shopping agents" that mimic legitimate ones to extract sensitive pricing, inventory, or payment data from merchants
2. Unauthorized Agent Purchases — Compromised or poisoned agents making purchases the human principal didn't authorize, leading to chargebacks
3. Agent Prompt Injection — Adversarial inputs that manipulate shopping agents into buying wrong products, accepting inflated prices, or bypassing spending limits
4. Agent-to-Agent Collusion — In B2B scenarios, agents from different parties could be manipulated to agree on harmful terms
5. Velocity Fraud — Autonomous agents can execute thousands of transactions per minute, overwhelming fraud detection systems designed for human-speed commerce
6. Identity Verification Gap — How does a merchant verify that an AI agent is truly authorized to act on behalf of a specific human/company?

Existing fraud prevention (Signifyd, Riskified, Forter) is designed for human purchasers. They analyze browser fingerprints, typing patterns, device IDs — none of which exist in agent transactions.