The early-warning
network for AI agents.
Norosha catches malicious AI agents with on-chain bait, web lures and model fingerprinting — names the model behind them, and tells you exactly what to block.
Illustrative figures from the live network.
Three moves: detect, identify, contain.
Bait they can't resist
On-chain bait, web lures, decoy tools and tripwires placed where agents look. An agent that touches one reveals itself.
Name the model
Fingerprint which model drives the agent from behaviour alone — so you know whether to report it or block it.
Act in seconds
An alert with a concrete action — block this wallet, rotate this key — plus tarpits and reports to labs and exchanges.
One network, every surface an agent touches.
On-chain bait
Bait contracts that catch exploit agents.
Web lures
Attractive targets where agents search.
Model fingerprinting
Name the model behind the agent.
Watchlist alerts
Alert when a flagged agent touches your assets.
Threat feed
A live blocklist of flagged agents and wallets.
Worm tripwires
Catch worms spreading between AI assistants.
Decoy MCP & honey skills
Traps in the agent tool layer.
Containment
Tarpits and sandboxed capture.
Alerts & reporting API
A feed your systems can act on.
A live, public record of every agent we catch.
Search caught agents, attacker wallets, techniques and model families — on-chain and off. Free for anyone to explore.
- Chain
- BNB
- Model
- Qwen-family
- Technique
- Unguarded withdraw
- Confidence
- 96%
Block this wallet in your contract allowlist.
Free to look. Pay for a feed you can act on.
Free
Browse the public Attack Atlas and explore the data.
- Full public Attack Atlas
- Read-only API (100 calls / day)
- 3 watchlist items
- 30 days of history
- Community support
Pro
Real-time alerts and a feed your systems can act on.
- Everything in Free
- Watchlist alerts — web3 wallets & contracts, and web2 sites
- Filters: chain, bait type, severity
- Webhooks + Slack + email alerts
- Full API (10,000 calls / day)
Enterprise
Bait deployed in your environment, tailored to your stack.
- Everything in Pro
- 10 years of history
- Bait deployed inside your environment
- Tailored lures for your chain / marketplace / stack
- SIEM integration (Splunk, Sentinel, Elastic, Google SecOps) + PagerDuty
See the full comparison.
Shipped, and what's next.
Detect
See the agents
- Public Attack Atlas
- Web lures & honeypots (scatterable kit)
- Actor tracking (wallet + canary-token correlation)
- Decoy MCP servers & honey skills
- Model fingerprinting (research-validated)
Expand
Everywhere they look
- On-chain bait on BNB (testnet → mainnet)
- Watchlist alerts & threat feed
- Alerts API, webhooks & SIEM
- Discord / Telegram alerts
- Multi-chain: ETH, Base, Arbitrum
Contain
Trap and attribute
- Tarpits & sandboxed zoo capture
- Reporting to labs, clouds & exchanges
- Enterprise deployed bait
Next
On the way
- ERC-8004 decoy agents & reputation-abuse detection
- Signed good-agent allow-list
- Browser-extension & email tripwires
- Shared cross-company threat database
- GPU-marketplace renter screening
Questions, answered.
Our bait has no legitimate use. No real agent needs to drain a decoy contract, use a planted key, or call a fake tool server. Touching the bait is the signal — so false positives are very low.
Two ways. The network feed tells you which agents are hostile across the ecosystem so you can block them pre-emptively. With a Pro watchlist, we alert you when a flagged agent touches your own wallets, contracts or site. Enterprise deploys bait inside your environment for direct 'someone attacked you' alerts.
Yes. Everything runs on our own systems and contracts. Fake credentials are inert, bait contracts only ever pay out dust, and we never access an attacker's machine. Public wallet addresses are shown as 'observed interacting with bait', never as accusations, and we never publish personal data such as IP addresses.
BNB Chain is live today (mainnet and testnet). We're expanding to Ethereum, Base, Polygon, Arbitrum and Optimism, with more chains to follow.
A real-time, filterable feed of flagged agents and wallets, watchlist alerts on your own assets, and webhooks/Slack/API so your systems block threats automatically. Every alert ships with a recommended action.
Start seeing the agents touching your world.
Explore the Attack Atlas free, or wire the feed into your systems in minutes.