Hi DEV Community! 👋
I wanted to share a project I've been working on for the Colosseum Hackathon called SolSecureAI.
With the explosive rise of Large Language Models (LLMs), production vulnerabilities like prompt injection are becoming critical security vectors. Traditional automated scanners often miss these flaws because security auditing fundamentally requires human ingenuity.
To solve this, I built a crowdsourced AI red-teaming platform that leverages blockchain smart contracts to manage security rewards transparently.
🛠️ The Tech Stack
• Smart Contracts: Written in Rust using the Anchor framework to handle fund escrows and on-chain challenge verification.
• Backend: Node.js managing isolated proxy architecture and live WebSocket sessions to route prompt payloads into target LLMs.
• Frontend: Next.js optimized for a smooth, real-time playground interface.
• Deployment: Hosted seamlessly via Railway.
🏗️ How the Architecture Works
- Challenge Launch: Companies set up an evaluation challenge, defining core system guidelines and funding an on-chain Solana escrow contract.
- Red-Teaming Playground: Security researchers craft real-time injection payloads directly within an interactive web console.
- Automated Validation: If a user successfully triggers a verifiable system-instruction bypass or exfiltration, our validation layer communicates with the smart contract to instantly release the bounty directly to the researcher’s Web3 wallet.
🚀 Try It & Open Source
The platform is currently live on Solana Devnet, and the entire repository is fully open-source because I believe security tools should be verifiable by the community.
• Live Demo: https://solsecureai.up.railway.app
• GitHub Repository: github.com
I would love to get your feedback on the verification model, the Node-to-Rust communication architecture, or any suggestions you have for expanding decentralized AI auditing!
Top comments (0)