
AI agents can help businesses manage money. But should an AI agent have the power to move company funds on its own? I don't think so. That's why I built VaultGuard for the @NEARProtocol Hackathon. Here's what it does 🧵

DeFi. Full Stack Founder: @G_WNXT

AI agents can help businesses manage money. But should an AI agent have the power to move company funds on its own? I don't think so. That's why I built VaultGuard for the @NEARProtocol Hackathon. Here's what it does 🧵

Live Demo URL: vaul-guard.guadgodiliamalini… GitHub Repository: github.com/danythost/Vault-g…

9/9 – The bigger picture The goal isn't to give AI more control over money. It's to make AI-powered financial automation safer. AI recommends. NEAR smart contracts authorize. The payment executor executes. This is VaultGuard, built for the @NEARProtocol Hackathon. #NEARHackathon

8/9 – Security testing We built a Security Center with 42 adversarial security checks covering: Treasury custody attacks. Attestation and TEE spoofing. Cross-chain settlement attacks. Our reported results: 42/42 passing security gates.

7/9 – Cross-chain payments Once a payment passes the required checks, VaultGuard can use NEAR Intents 1Click to facilitate payments across supported blockchains. Businesses can pay vendors in supported tokens without manually managing every blockchain transaction.

6/9 – What makes VaultGuard different? → AI-powered invoice analysis. → Restricted NEAR Function Call Keys. → Onchain spending limits and vendor controls. → Cryptographic attestations. → Cross-chain settlement through NEAR Intents 1Click. → Auditable payment activity.

5/9 What happens during an attack? Imagine someone hides instructions in an invoice telling the AI to send $10,000 to an attacker's wallet. Even if the AI recommends approval, the smart contract checks the actual payment details. If the rules are broken, the payment is rejected.

4/9 How it works → A business submits an invoice. → AI analyses it for suspicious information. → Vendor and recipient details are checked. → Spending limits and company policies are evaluated. → The NEAR smart contract independently checks whether the payment is authorized.

3/9 – The idea VaultGuard follows one simple principle: Let AI do the thinking, but never give it control of the money. AI analyses invoices and identifies risks. NEAR smart contracts independently enforce the rules before a payment can proceed.3/9 – The idea

2/9 – The problem Businesses process hundreds of invoices and payments. Using AI can make this faster, but giving an AI agent direct access to company funds is risky. One malicious invoice or prompt injection could lead to an unauthorized payment.

Proposal alert! This one proposes a reusable Gimbalabs Brand Kit: logos, fonts, colors, presentation templates, and animations for Open Spaces videos. Making it easier for anyone to represent and contribute to Gimbalabs! github.com/gimbalabs/season7… #FeedbackStage #GimbalabsOpenGovernance

Gimbalabs has always been more than one team. Educators, SPOs, developers, Catalyst projects, and community organizers have all helped shape our work. Have we worked together? Should we? We're reconnecting with our network. Come say hello at Open Spaces! #GimbalabsUpdates



