Loops HouseBounTEE

NEAR HACKATHON
Team
1 member- LIOwner
Liam Pereira
Overview
BounTEE: bug bounties that pay themselves. A Whitehat, or their AI agent, seals an exploit and a TEE replays it. If the exploit is proven, NEAR pays them and signs a pause on every chain the bounty covers.
The problem: Protocols spend time and money on bug reports that can't be reproduced, and AI now writes them for free. When a report is real, the bug stays live on every chain until someone reads it and acts. Whitehats hand over the exploit before they're paid, so severity gets downgraded and payment delayed.
How it works: a Protocol posts a USDC bounty on a NEAR contract. The Whitehat seals the exploit on their own machine and files a claim. BounTEE's judge replays it in a TEE (Intel TDX on OutLayer) against the chain's recent state. If it steals at least the Protocol's minimum from the covered contracts, the contract pauses them on every chain and pays the Whitehat the whole bounty. Only then can the Protocol open the exploit.
How NEAR fits, all run on mainnet (links in the README):
- A keyless contract, bounty3.bountee.near, holds the USDC. Nobody, including us, can take it or block a valid payout.
- Chain Signatures sign the pause on Base and Arbitrum One from the bounty's own address.
- MPC foreign-chain verification confirms each pause really landed.
- A function-call access key lets an AI agent claim without being able to move the Whitehat's money.
- OutLayer keeps the exploit sealed: a copy sent to OutLayer directly is refused.
The mainnet demo: a cheat claim paid nothing, an AI agent's real claim was paid 10 USDC, and both covered vaults were paused. Anyone can check every claim with no keys and no funds.
Live site: bountee.dev. Try it on testnet: testnet.bountee.dev. Whitehat walkthrough on testnet (video): https://youtu.be/8SQj71whDwc