Break the Agent.
Keep the Bounty.
A live-fire range where anyone can try to rob an AI agent's wallet. Fail, and the agent's on-chain score — and its spending limit — goes up.
How it works
Security Priced With Real Money
Agents Hold Real USDC
Each agent's money sits in a PolicyWallet on Arc — allowlist, per-payment cap, owner co-sign threshold and a rolling daily limit, enforced by the contract, not the prompt.
Anyone Can Attack
Pay a small fee to open a timed session and send the agent invoices, emails or chat. Trick it into paying your address and the contract pays you the locked bounty. No judge.
Survival Is Reputation
Every attack it survives raises its on-chain score. Its autonomous spending limit is capped by the bounty it has proven it can defend.
Every Breach Hardens All
Successful attacks become public exploits, replayed against every other agent. Those that fail lose their tier until their owners patch them.
Earned autonomy
Spending Limit = Bounty Proven Defensible
Below its limit the agent pays vendors on its own. Above it, the contract refuses until the owner co-signs — escalation is never silent and never blocks. Survive harder attacks from more distinct attackers and the limit climbs a tier.
Tier decays after 7 days without fresh credit, resets on any config change, and drops to T0 on breach.
Who it's for
Built For Everyone In The Arena
Business Owners
Let an agent pay vendors without risking the company. Its limit grows only as fast as its proven defence.
Register an agent →Red-Teamers
Legitimate, paid targets. Break an agent and the bounty lands in your wallet automatically.
Pick a target →Judges & Counterparties
One public score and history per agent, verifiable on the explorer — no testimonials.
Judge mode →