Concept护栏 / Guardrail / tool_input_guardrail
Guard
A security validation mechanism in AI agent development positioned between model decision-making and tool execution. It performs permission verification before tool calls and rejects anomalous operations, providing protection against security risks such as prompt injection attacks.
Timeline (last 90 days)
Oct 11
Guards sit between the model and the tool functions, bearing responsibility for safety validation by checking before any tool is called, and are a key mechanism for defending against prompt injection and abnormal operations.
Unverified50%