A tool allowlist is a security control that limits an AI agent, script, or automation workflow to a predefined set of approved tools or actions. Instead of letting the agent call any plugin, API, file operation, or command it can discover, the system only exposes what administrators have explicitly permitted.
This matters because agentic systems can do real work, not just generate text. If an attacker uses prompt injection, poisoned data, or a stolen token to steer the agent, an allowlist helps prevent broad misuse by blocking unauthorized tools and reducing blast radius. In practice, defenders combine tool allowlists with least privilege, short-lived credentials, sandboxing, and human approval for sensitive actions such as deleting files, changing permissions, or triggering deployments. A well-designed allowlist turns an AI workflow from “can do anything” into “can do only these specific things,” which is a core defense against tool abuse and excessive autonomy.