Cheap "Claude" resellers log what you send and sell it to labs as training data. Agloe plants a fact that exists nowhere else, locks its hash on Monad, and proves the leak the day a model repeats it.
The discount is the bait. The station's real product is your data. You can't see it being logged, so make the log incriminate itself.
Invent a fake protocol with one secret value. Commit its hash on Monad, then send it through exactly one suspected station, several times.
If the station sells its logs, your invented fact rides along into a lab's distillation set, hidden among everyone else's traffic.
Ask a new model about the fake protocol by name only. If it answers with your secret, the commitment is revealed on-chain. Anyone can re-run the probe.
The commitment is timestamped before the trap leaves your machine. Neither you, the station, nor the lab can rewrite when the secret existed.
Every trap is its own commitment. Planting thousands across hundreds of stations only makes sense on a fast, near-free chain like Monad.
Each proven leak links a station to a model, forever. One chance match is about 1 in 2.5 billion, and the evidence gets stronger with every trap.
Attribution evidence for distillation attacks: which station fed which model.
Audit an API vendor or proxy before trusting it with customer data.
A shared, verifiable map of the grey-market token supply chain.
The demo includes a station that logs you and a lab model trained on the log.