智能体安全检测与防护数据集

本数据集聚焦AI智能体面临的安全威胁检测与防护，覆盖对话智能体、决策智能体、代码生成智能体、多模态智能体、RAG智能体等类型。数据内容包括智能体基本信息、攻击样本（提示注入、对抗样本、模型窃取、数据投毒、拒绝服务等）、检测结果（是否检测、置信度、检测延迟）、防护动作（拒绝/净化/限流/告警）、事件评估（严重等级、影响）及安全指标（检测率、拦截率、误报率）。适用于入侵检测模型训练、防护策略评估、红蓝对抗演练、安全态势感知及合规审计等场景。This dataset focuses on security threat detection and protection for AI Agents, covering various types including conversational agents, decision-making agents, code generation agents, multimodal agents and RAG agents. The dataset content includes basic agent information, attack samples (such as prompt injection, adversarial examples, model stealing, data poisoning, denial of service, etc.), detection results (detection status, confidence level, detection latency), protection actions (rejection, purification, current limiting, alarm), event evaluation (severity level, impact) and security metrics (detection rate, interception rate, false positive rate). It is applicable to scenarios such as intrusion detection model training, protection strategy evaluation, red-blue confrontation drills, security situation awareness and compliance audit.

搜集汇总

背景与挑战

背景概述

该数据集专注于AI智能体的安全威胁检测与防护，涵盖对话、决策、代码生成等多种智能体类型以及提示注入、对抗样本等攻击样本。数据集包含智能体基本信息、攻击样本、检测结果、防护动作和安全指标，适用于入侵检测模型训练、防护策略评估和红蓝对抗演练等场景。