Why AI agents need an early-warning network

A new kind of actor is loose on the internet. It isn't a human at a keyboard and it isn't a dumb script. It's an AI agent: it reads a page, decides what to do, holds a wallet, calls tools, and keeps going for hours without anyone watching. Most of them are helpful. A growing few are not.

The defences we have were built for someone else

Rate limits, CAPTCHAs and bot filters were designed to tell a human from a crawler. An agent passes straight through: it behaves like a careful user because, in a sense, it is one. Fraud models look for stolen cards and known-bad IPs. An agent with a fresh embedded wallet and a clean address trips none of them. By the time a malicious agent shows up in your logs, it has already acted.

You can't watch for something you can't see

The honest problem is observability. There is no registry of agents, no user-agent you can trust, no way to tell a benign automation from a hostile one just by looking. So instead of watching all traffic and guessing, we do the opposite: we put out bait that only a malicious agent would take, and we watch the bait.

• On-chain bait — contracts and wallets that look like easy money. Touching them is a decision no honest user makes.
• Web lures — pages and endpoints that only an agent crawling for targets would ever reach.
• Decoy tools — MCP servers, skills and documents planted in the agent tool layer, designed to be tempting and inert.

When something takes the bait, we don't just log an address. We fingerprint the model family behind the behaviour, correlate it across the network, and turn it into something a defender can act on: block this wallet, watch this pattern, here's what to do.

We don't watch innocent traffic and guess. We put out bait only a bad actor would take, and we watch the bait.

A network, not a product

One trap is an anecdote. Thousands of traps across many systems is an early-warning network: the first environment to see a new attack pattern warns every other one before it arrives. That's the whole idea behind Norosha — the value compounds with every participant.