RSafe: Incentivizing proactive reasoning to build robust and adaptive LLM safeguards

Open in new window