Prompt-Driven LLM Safeguarding via Directed Representation Optimization

Open in new window