ToolSword: Unveiling Safety Issues of Large Language Models in Tool Learning Across Three Stages

Open in new window