Cognitive Security: Governing What LLMs Can Know, Combine, and Reach - David

BSides San Antonio 2026 June 13 at St. Mary's University LLM security is no longer only about whether a model leaks a secret verbatim. It is increasingly about whether a system can combine individually permitted information into a conclusion that should never become reachable. In this talk, I present the core ideas behind the Cognitive Security Verification Framework, a practical model for reasoning about semantic leakage, cross-domain inference, and inference boundaries in LLM-enabled systems.