MOI is engineered from first principles under the assumption that software agents and external RPC endpoints may be untrusted, compromised, or adversarial. Capital is safeguarded through a layered defense-in-depth model.
Threat Models & Architectural Mitigations#
| Threat Vector | Potential Impact | MOI Architectural Defense |
|---|---|---|
| Agent Compromise / Prompt Injection | Unauthorized capital transfer to attacker | Restricted signer cannot transfer funds; target contracts restricted to 0xa9AA... |
| Front-Running / Sandwich Attacks | Slippage loss on decentralized exchanges | Textile RFQ firm quotes execute at guaranteed rateRay via direct limit reactor |
| Replay / Double-Spend Attacks | Duplicate execution of a valid quote | Single-use nonces, Permit2 expiration, and database uniqueness locks |
| Unauthorized Initial Execution | Agent begins trading before owner review | Rule #2: Mandatory First-Trade Owner Authorization Gate |
| Cross-Tenant Data Leakage | Tenant B reads or executes against Tenant A treasury | EIP-4361 SIWE session with PostgreSQL Row-Level Security (RLS) |
| Sudden Market Dislocation | Depleting reserves in volatile devaluations | Dynamic circuit breaker tripping on cumulative daily loss |