An early-warning network for the agent era.

AI agents can now act on their own — browsing, running code, moving funds. Most of the time that's useful. Sometimes it isn't: agents are already being used to probe systems, exploit smart contracts, and spread between tools. The people defending against them are mostly flying blind.

Norosha gives them sight. We place bait where agents look — on-chain, on the open web, and in the agent tool layer — and when something takes it, we name the model behind it and tell defenders exactly what to do. One network, many sensors, feeding a single picture of what's out there.

We started with research: measuring how real AI models behave against bait, how fast a self-copying agent could spread, and how reliably a model can be identified from its behaviour alone. That evidence is the foundation everything here is built on.

Our mission is simple: make the agent era safe to build in, by catching the agents that shouldn't be trusted before they do harm.