Field notes.
Research, thinking and lessons from building the early-warning network for AI agents.
How malicious AI agents attack web3 — and how to catch them
Crypto is the perfect hunting ground for hostile AI agents: programmable money, irreversible transactions, and wallets an agent can control directly. Here's what the attacks look like, and how on-chain bait catches them.
ReadDetecting malicious AI agents: bait, fingerprinting and actor tracking
You can't watch for something you can't see. Here are the three signals that actually work for catching hostile AI agents — and how they combine into attribution a defender can act on.
ReadWhy AI agents need an early-warning network
Autonomous agents now hold wallets, call tools and act for hours unsupervised. The defences built for humans and for bots don't see them. Here's the gap — and how bait closes it.
ReadWhat we learned baiting frontier models
We ran controlled experiments against today's leading models. Lures work better than anyone expects — with one striking exception — and the way you phrase a trap matters more than the trap itself.
Read