agentic commerce

The Zero-Signal Problem: Why Traditional Fraud Detection Fails for AI Agents

F
Faultr Team
March 5, 20268 min read

Fraud detection today is built on a mountain of behavioral signals: mouse movements, keystroke rhythms, browsing history, and IP geolocations.

When an AI agent makes a purchase, all of these signals disappear. This is the Zero-Signal Problem.

What is the Zero-Signal Problem?#

Traditional fraud detection engines are designed to distinguish between a "legitimate human" and a "bot." In the era of agentic commerce, the bot is the legitimate customer.

The agent doesn't have a mouse. It doesn't "browse" the page. It makes API requests or interacts with a headless browser in a perfectly efficient, non-human way. If your fraud engine flags non-human behavior, it will flag 100% of your most valuable agentic traffic.

The Behavioral Void#

  • No Browser Fingerprinting: Agents often use standardized headless environments.
  • No Biometrics: Keystroke dynamics and mouse heatmaps are non-existent.
  • Perfect Efficiency: Agents go straight from intent to transaction in milliseconds.
The Fraud Risk

Without behavioral signals, a malicious agent looks identical to a legitimate one. This opens the door for large-scale, automated sophisticated fraud that circumvents legacy defenses.

What Replaces Behavioral Signals?#

In the absence of human-centric signals, we must move toward Protocol-Centric Compliance.

Instead of asking "Is this a human?", we must ask:

  1. "Does this agent follow the AP2 (Agent Payment Protocol) standard?"
  2. "Is the intent consistent with the agent's historical evaluation profile?"
  3. "Does the transaction meet the verifiable constraints of the merchant's policy?"

Introducing AP2 Compliance#

The Agent Payment Protocol (AP2) is designed to include cryptographically signed intent and verifiable constraints.

json
{
  "intent": "purchase",
  "item_id": "sku_9921",
  "max_price": {
    "amount": 29.99,
    "currency": "USD"
  },
  "signature": "0x7f...ac4",
  "protocol_version": "AP2-v1.2"
}

At Faultr, we help merchants and agent-builders bridge this gap by providing adversarial testing for these new protocol-driven signals.

In the next part of this series, we'll explore Intent Verification and how it replaces the traditional "browsing" signal.

Share