Our work on reinforcement fine-tuning-based red teaming benchmark is accepted by Frontiers of Computer Science.