LLM policy channel security research — unauthenticated injection, cryptographic fix, threat model. The question nobody was asking.
Security research prototype testing unauthenticated policy channels in LLM architectures. Identified vulnerability class (CWE-306), built threat model, developed cryptographic signing fix, and documented responsible disclosure framework.