Stop your AI agent from paying a scammer

Описание и применение

Автоматический рецепт защиты AI‑агентов от отправки средств мошенникам с помощью verdix-guard, проверяющего адреса перед подписанием.

Задача

AI agents can unknowingly transfer funds to malicious or sanctioned addresses due to prompt injection, address poisoning, or lack of pre‑signature checks.

Результат

Payments are halted before signing when the recipient address is flagged as risky or sanctioned, preventing loss of funds to malicious actors.

Рабочий процесс

• 1. Install verdix-guard and dependencies via npm.
• 2. Import and create a Verdix guard with the agent’s account.
• 3. Wrap the x402 client or viem account with the guard.
• 4. Execute payment or transaction; the guard checks the recipient address via Verdix API before signing.
• 5. If the address is flagged, the guard throws an error and the transaction is aborted.

Доказательства

An LLM decides what to do from text, and some of that text comes from people you don't control: web pages, emails, tool results, other agents.

Источник

Addresses are 40 hex characters, and people (and agents) copy them from transaction history. An attacker generates a lookalike address with the same first and last characters as one you've used before, then sends you a tiny or zero-value transfer from it, so it shows up in your history.

Источник

The check has to run in the signing path: outside the model, on the final recipient, failing closed.

Источник

verdix-guard hooks into two places: x402 payments and any transfer via viem.

Источник