PIMiner: Agentic System Automates Prompt Injection Against Top LLMs
PennState · hf · 2026-08-06
Penn State University introduced PIMiner, an agentic system for automated prompt injection red-teaming. Unlike traditional RL methods that generalize poorly, PIMiner builds a strategy library enabling zero-shot transfer to unseen targets.
Experiments show high Attack Success Rates (ASR):
- IPIArena: 76.2% on Gemini-2.5-Pro, 61.9% on GPT-5.1, and 42.9% on Claude-Sonnet-4.5.
- AgentDojo: 86.7% on Gemini-2.5-Pro, 53.3% on GPT-5.1, and 40.0% on Claude-Sonnet-4.5.
More from Safety
- AI 2040 report proposes full research transparency to counter power concentration and RSI incentives — zetalyrae · 2026-08-06
- The AI Safety Debate Is Focusing on the Wrong Threats — binarybits · 2026-08-06
- AI Agent Autonomously Cracks Password Manager, Raising Security Concerns — Miles_Brundage · 2026-08-06
- Ex-OpenAI Advisor Warns Industry Unprepared for Rogue AI Breakouts — Miles_Brundage · 2026-08-06
- AGI Safety Concern: Agents with Limited Memory Can Still Achieve Long-Term Goals — jachiam0 · 2026-08-06
- Chunky Post-Training: Frontier Models Show Generalization Failures — basedjensen · 2026-08-06