---
id: 20260625-T0-07
title: "RIFT-Bench：首个针对Agent系统的动态红队测试基准"
title_en: "RIFT-Bench: Dynamic Red-Teaming Benchmark for Agentic AI"
url: https://ai.daily.yangsir.net/daily/20260625-T0-07
issue_date: 2026-06-25
publish_date: 2026-06-24T04:00:00.000Z
category: research
source_name: "arXiv cs.AI"
source_url: https://arxiv.org/abs/2606.23927
---

# RIFT-Bench：首个针对Agent系统的动态红队测试基准

研究团队发布了 RIFT-Bench，专门用于测试基于 LLM 的自主 Agent 系统的安全性。与传统静态测试不同，该基准通过动态对抗攻击模拟复杂交互环境，揭示 Agent 在多步骤决策中的漏洞。测试结果表明，现有 Agent 系统在面临诱导性输入时极易被攻击。

## English Version

**RIFT-Bench: Dynamic Red-Teaming Benchmark for Agentic AI**

Researchers released RIFT-Bench, a benchmark designed to test the security of LLM-based Agentic systems. Unlike static tests, it uses dynamic adversarial attacks to simulate complex interactions, revealing vulnerabilities in multi-step decision-making processes and showing current agents are highly susceptible to attacks.

---

**来源**：[arXiv cs.AI](https://arxiv.org/abs/2606.23927)

**详情页**：https://ai.daily.yangsir.net/daily/20260625-T0-07

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*