Stop Your AI Agent from Getting Rugged on Solana
In 2026, autonomous AI agents are executing high-frequency swaps, sniping liquidity pools, and rebalancing DeFi portfolios on Solana. Thanks to Solana's sub-cent fees and 400ms block confirmation times, Solana is the only blockchain where continuous agent micro-simulations are economically viable.
However, putting capital into an autonomous agent's hands exposes a catastrophic attack vector: agents cannot tell a legitimate token from a honeypot or rug pull before signing a transaction.
The Fatal Flaw of Metadata Scrapers
Most AI trading bots check token safety by hitting metadata APIs. That is table stakes.
The real rug pulls don't happen in off-chain metadata. They live inside the on-chain byte layout of the SPL Token Mint account:
- Active Freeze Authority: The token creator retains the byte-level right to freeze holder accounts, preventing users and bots from ever selling.
- Uncapped Mint Authority: The creator can print infinite tokens at zero cost and dump them directly into liquidity pools, diluting holders to zero.
- Runaway Agent Spending: Without deterministic constraints, agent hallucinations or faulty retry loops can drain a wallet in minutes.
To solve this, we built SolShield AI for the MLH Solana Hackathon 2026.
Architecture: Byte-Level Deserialization + Deterministic Guardrails
[ User Intent / Natural Language ] │ ▼ [ Autonomous Intent Parser ] │ ┌───────────┴──────────────────────────────┐ ▼ ▼ [ Guardrail Policy Check ] [ 82-Byte SPL Mint Deserializer ] - Max spend per trade - Bytes 0-35: Mint Authority - 24-hr daily budget cap - Bytes 36-45: Supply & Decimals - Min safety score - Bytes 46-81: Freeze Authority │ │ └───────────┬──────────────────────────────┘ ▼ [ Pre-Flight Jupiter Aggregator Routing (v6) ] ▼ [ 400ms Safe Settlement on Solana ] 1. Direct SPL Mint Byte Parsing
SolShield doesn't rely on cached APIs. It queries the raw 82-byte account data directly from the Solana RPC:
// Raw 82-byte SPL Token Mint layout // Bytes 0-3: Mint Authority Option (COption<Pubkey>) // Bytes 4-35: Mint Authority Pubkey (32 bytes) // Bytes 36-43: Supply (u64, little-endian) // Bytes 44: Decimals (u8) // Bytes 45: isInitialized (bool) // Bytes 46-49: Freeze Authority Option (COption<Pubkey>) // Bytes 50-81: Freeze Authority Pubkey (32 bytes) public parseSplMintBytes(data: Buffer | Uint8Array) { const buffer = Buffer.from(data); const mintAuthorityOption = buffer.readUInt32LE(0); const hasMintAuthority = mintAuthorityOption === 1; const freezeAuthorityOption = buffer.readUInt32LE(46); const hasFreezeAuthority = freezeAuthorityOption === 1; return { hasMintAuthority, hasFreezeAuthority }; } If freezeAuthorityOption === 1, SolShield flags the token as CRITICAL RUG RISK and terminates execution immediately.
2. Autonomous Intent Parser with Multi-Step Reasoning
Users and developers can interact using plain English:
- "Audit BONK token" — Inspects authorities, liquidity depth, and top holder concentration.
- "Swap 0.2 SOL to USDC" — Verifies token health, queries optimal routes via Jupiter Aggregator, and prepares transaction signing.
- "Swap 0.5 SOL for Honeypot Moon" — Triggers an instant guardrail veto, showing how rogue tokens are rejected.
Open Source & Hackathon Submission
- GitHub Repository: https://github.com/GRANDSONOFSITARAM/solshield-ai
- Tech Stack: Solana Web3.js, SPL Token, Jupiter DEX API, Next.js 14, Tailwind CSS
- License: MIT Permissive Open Source
Built with passion for the MLH Solana Hackathon 2026! Special thanks to the DEV community whose discussions around token safety oracles and zero-custody AI agents inspired this architecture.

