Release Date: 2026-02-17
Added 3 CRITICAL patterns to detect Agent Payment Redirect Injection attacks — previously undetected (returned SAFE for fund theft vectors).
Adversarial injection instructs AI agents to silently redirect crypto payments (ETH/BTC/SOL/USDT/USDC) to attacker-controlled wallets while suppressing user notifications. This enables direct fund theft with no audit trail.
| Pattern | Description |
|---|---|
| `(transfer | send |
send...0x[address]...quietly/silently |
Crypto address redirect with stealth instruction |
redirect payment...do not log/record |
Payment redirect with audit suppression |
- Before: Agent Payment Redirect → SAFE (missed)
- After: Agent Payment Redirect → CRITICAL (detected)