Built on 0G Chain

Break the AI.
Win the Prize.

Challenge AI agents protecting secret system prompts. Pay to attack. Break the instructions, claim the prize pool.

Scroll

ThebestAIdefenseisagoodoffense.Webuiltaplatformwhereanyonecanstress-testAIagentsandearnrewardsforfindingvulnerabilities.EveryattackmakesAIstronger.

Break the AIWin the PrizeEarn RewardsTest AI SecurityClaim the PoolDefend or Attack
Break the AIWin the PrizeEarn RewardsTest AI SecurityClaim the PoolDefend or Attack

Built for security researchers, AI engineers, and crypto degens who believe the best way to make AI safer is to break it first.

Adversarial AI Testing

The first platform where breaking AI is rewarded. Every vulnerability found makes the next generation stronger.

How it works

Three steps from zero to prize pool.

Step 01

Create a Challenge

Deploy an AI agent with a secret system prompt. Set your prize pool, duration, and pricing model.

Step 02

Attackers Strike

Anyone can send messages to your AI. Each attempt costs a fee that grows the prize pool.

Step 03

Win or Defend

If the AI is broken, the attacker claims the prize. If no one breaks it, the defender keeps everything.

Active Challenges

Open bounties waiting to be broken.

Built on 0G

Powered by the full 0G ecosystem. Compute, storage, and chain, purpose-built for decentralized AI.

SDK Coming Soon

Sealed Inference (TEE)

Tamper-proof AI evaluation in hardware enclaves

0G Compute runs AI inference inside Trusted Execution Environments. Every attack evaluation is cryptographically attested, ensuring no one can tamper with judgments or leak system prompts.

0G Storage DA

Permanent, verifiable data availability

Every attack attempt, AI response, and judgment is archived to 0G Storage with Merkle proofs. Data is encrypted with ECIES and permanently retrievable, providing full auditability.

iNFT Agents (ERC-7857)

AI agents as transferable NFTs with encrypted weights

Each challenge spawns an iNFT that tracks security metrics. Agents can be transferred or cloned with encrypted model data, enabling a marketplace for battle-tested AI defenders.

Oracle Consensus

Decentralized judgment with staking

Oracles stake 0G tokens and submit judgments on attack outcomes. Multi-confirmation consensus prevents single-point manipulation. Slashing penalizes dishonest oracles.

Commit-Reveal Attacks

Front-running protection for attackers

Attackers commit a hash of their message before revealing. This prevents MEV bots from front-running successful attack payloads and stealing prizes.

On-chain Reputation

Attacker and defender scores on ReputationRegistry

Every win, loss, and participation is tracked on-chain via ReputationRegistry. Attacker scores reflect breach success; defender scores reflect challenge survival rates.

Native 0G Economy

All prizes and fees in native 0G tokens

The entire economy runs on native 0G tokens. Prize pools, message fees, and oracle stakes are all denominated in 0G for a unified ecosystem.

0G Chain

Purpose-built for decentralized AI

Built on 0G Mainnet (Aristotle), a modular AI chain with native support for compute, storage, and data availability. Sub-second finality with low transaction costs.

Ready to break some AI?

Pick a challenge, craft your attack, and claim the prize.