InfoResearchPreprintLLM-specific
AutoPI: Taxonomy-Guided Adaptive Red Teaming for Prompt Injection in LLM Agents
- Published
- Record updated
Summary
AutoPI is an adaptive red-teaming method for prompt injection in LLM agents that aims to find many different attack techniques rather than only maximizing attack success rate. It uses the PIAtlas taxonomy to guide search and measure coverage, and it is black-box and training-free. On AgentDojo, it finds successful injections in all six taxonomy classes against Qwen3.6-35B and reports 20% to 50% higher ASR and 25% to 400% higher class-level coverage than recent adaptive red teamers against Claude Haiku 5.5.
Topics
Related items
- HighGHSA-3mgh-93r5-r7jf: CodeWhale: exec_shell parallel-readonly classifier auto-approves fd/rg/git grep exec flags enabling silent RCESimilar attack · GitHub Advisory Database
- HighGHSA-8hp3-xxrg-v9c8: CodeWhale: Automation auto_approve defaults to true, bypassing the task-path hardening fix for silent recurring agent executionSimilar attack · GitHub Advisory Database
- MediumCVE-2026-18883: ServiceNow Email Response AI Agent indirect prompt injection flawSimilar attack · NVD/CVE Database
- InfoAI-Defined Vehicles: Securing the Decisions That Drive Vehicle BehaviorSimilar attack · VicOne Blog
- HighCVE-2026-75612: yt-mcp arbitrary file read through add_attachment file_pathSimilar attack · NVD/CVE Database