Specialized Models Beat Single LLMs for AI Security | AI Sec Watch