Learning When to Act or Refuse: Guarding Agentic Reasoning Models for Safe Multi-Step Tool Use Paper • 2603.03205 • Published Mar 3 • 13
