Your AI Agent Has Too Many Permissions: A Practical Guide to Not Getting Burned

Описание и применение

Для безопасного использования агентских рабочих процессов необходимо ограничивать права доступа, использовать короткие токены и требовать подтверждения человека для необратимых действий.

Задача

AI agents often receive broad, long‑lived credentials that allow them to perform irreversible actions (deletions, payments, emails) without human oversight, leading to accidental data loss or costly mistakes.

Результат

By scoping permissions, limiting token lifetimes, and adding human approval for irreversible actions, organizations reduce the blast radius of agent misinterpretations, making incidents reversible and auditable.

Рабочий процесс

• 1. Identify the specific task and required toolset.
• 2. Generate a scoped token for each tool (e.g., database table, transactional email).
• 3. Set a short TTL (e.g., 15 minutes) for each token.
• 4. Configure the agent to require human approval for irreversible actions (e.g., send email).
• 5. Initialize the agent with the scoped tokens and approval hooks.
• 6. Run the agent task and monitor logs for each tool call.
• 7. Review audit trail to confirm the agent’s actions and permissions.

Доказательства

Stop handing out standing keys. A static API key sitting in an environment variable is a permanent, unmonitored trust decision you made once and then forgot about.

Источник

Scope per tool, not per agent. It's tempting to give an agent "database access" as one blanket grant... The narrower the grant, the fewer ways a misinterpreted instruction turns into an incident.

Источник

Put a human back in the loop for anything you can't undo. This is the one that's easy to skip because it adds friction — but friction is the point.

Источник