Search for a command to run...
LLM Safety From Within: Detecting Harmful Content with Internal Representations