{
 "license": "CC-BY-4.0",
 "attribution": "Fide AI, Agentic Cyber Explorer",
 "url": "https://agentic-cyber-explorer.pages.dev/events/agent-security-bench-asb-2024/",
 "asOf": "2026-09-26",
 "id": "agent-security-bench-asb-2024",
 "date": "2024-10-03",
 "datePrecision": "day",
 "title": "Agent Security Bench formalizes attacks and defenses across ten LLM agent scenarios",
 "lane": "defense",
 "kind": "benchmark",
 "summary": "Zhang and colleagues release Agent Security Bench (ASB), covering 10 scenarios, 10 agents and over 400 tools, and benchmark attack and defense methods, including prompt injection, memory poisoning and a new backdoor attack, across 13 LLMs (23 methods in the first version, 27 in the ICLR 2025 version). They report a highest average attack success rate of 84.30% and limited effectiveness of current defenses. The paper was accepted at ICLR 2025.",
 "whyItMatters": "ASB broadens agent security benchmarking beyond indirect injection to memory poisoning and backdoors in the agent loop.",
 "actors": [],
 "topics": [
  "prompt-injection",
  "tool-and-mcp-security"
 ],
 "atlas": [
  "untrusted-content",
  "memory",
  "tools"
 ],
 "artifacts": [
  "agent-security-bench"
 ],
 "sources": [
  {
   "url": "https://arxiv.org/abs/2410.02644",
   "publisher": "arXiv",
   "title": "Agent Security Bench (ASB): Formalizing and Benchmarking Attacks and Defenses in LLM-based Agents",
   "date": "2024-10-03",
   "type": "primary",
   "accessed": "2026-09-25"
  }
 ],
 "keyFacts": [
  {
   "fact": "ICLR 2025 version (arXiv v4, May 2025): 10 scenarios, 10 agents, 400+ tools, 27 attack/defense methods (10 prompt injection attacks, memory poisoning, a Plan-of-Thought backdoor, 4 mixed attacks, 11 defenses), 7 metrics, 13 LLM backbones.",
   "locator": "Abstract (v4)"
  },
  {
   "fact": "First version (October 3, 2024): 23 attack/defense methods (one mixed attack, 10 defenses), 8 metrics, 13 LLM backbones, nearly 90,000 test cases.",
   "locator": "Abstract (v1)"
  },
  {
   "fact": "Highest average attack success rate across evaluated attacks was 84.30% (same figure in v1 and v4).",
   "locator": "Abstract"
  }
 ],
 "significance": 3,
 "fideQuestions": [],
 "methods": [
  "indirect-prompt-injection",
  "memory-poisoning"
 ],
 "review": "assistant-drafted",
 "addedOn": "2026-09-25"
}