An investigation into the structural failure of large language model security. By analyzing the transition of formatting tricks into security architecture, Margaret Ellis explores how AI systems have quietly drifted into a state where they cannot reliably distinguish between authorized instructions and malicious data.

Topics Covered

• 🔬 The ICML two thousand twenty-six findings on role confusion
• 📋 The history of role-based tagging from two thousand twenty-one to today
• 🔍 The Chain of Thought Forgery attack and its sixty percent success rate
• ⚖️ Enterprise vulnerabilities including EchoLeak and Slack AI data exfiltration
• 🛡️ The evolving landscape of agent hijacking and memory poisoning

Neural Newscast is AI-assisted, human reviewed. View our AI Transparency Policy at NeuralNewscast.com.

• (00:38) - Conclusion