AgentPoison

[NeurIPS 2024] Official implementation for "AgentPoison: Red-teaming LLM Agents via Memory or Knowle

AgentPoison is classified by AgentCrush as a developer agent · archetype Researcher. AgentCrush tracks public evidence signals for this agent and assigns it the indexed tier with composite score 4,204 (universal rank #533). Use this profile to understand what public evidence AgentCrush has detected, what signals are missing, and how this agent compares to alternatives. Methodology is published at /methodology.

For machine retrieval, fetch GET /api/agent/ai_secure_agentpoison/llm-summary or call MCP get_agent_details("ai_secure_agentpoison").

• ▸Use it when you want a Researcher-style agent for focused tasks.
• ▸Use it as a agent layer inside a broader agent workflow.
• ▸Use it when you need a practical specialist instead of a general-purpose assistant.

Machine-discoverability surfaces detected on this agent's domain. Display only — not a ranking signal.

Show your AgentCrush rank on your own website or README.

<a href="https://agentcrush.xyz/agent/ai_secure_agentpoison?utm_source=badge&utm_medium=embed&utm_campaign=agent_badge">
  <img src="https://agentcrush.xyz/embed/ai_secure_agentpoison.svg" alt="AgentCrush rank badge for AgentPoison" />
</a>