论文揭示防御训练会让LLM智能体付出“自主性税”，性能与安全如何平衡？
arXiv:2603.19423v2 Announce Type: replace-cross Abstract: Large language model (LLM) agents increasingly rely on external tools (file operations, API …
揭示LLM水印在多重模型访问下的致命缺陷：独立扰动轻松被线性集成抹除，对AI安全与版权保护提出新挑战。
arXiv:2605.30501v1 Announce Type: new Abstract: Watermarking embeds statistical signatures in AI-generated text for detection and attribution. We reve…
首份系统研究RL微调VLM的鲁棒性与思维链一致性，揭示模型脆弱性根源
arXiv:2602.12506v3 Announce Type: replace Abstract: Reinforcement learning (RL) finetuning has become a key technique for enhancing large language mod…