Prompt Injection Defense Weekly

LoopTrap: How to stop an AI agent that won't quit

This week's attack: LoopTrap Termination Poisoning — semantically tricks LLM agents into infinite loops (86.3% success rate across GPT-4o, Claude, Gemini). Paired with a copy-paste Termination Integrity Guard (TIG) system prompt template to ship today.