Cross-Task Defense: Instruction-Tuning LLMs for Content Safety