Attack Success Rate (ASR) evaluates each jailbreak with a single yes/no label at the end of generation, telling us whether a failure happened but not how it unfolded. Two attacks that produce equally harmful outputs may have followed completely different paths, and ASR cannot tell them apart. We make those hidden paths observable from logits alone. Temporal Logit Observability (TLO) is a training-free diagnostic that watches a compliance-refusal margin during decoding and places each model-attack condition on a calibrated 2D plane. By design, this plane is most informative exactly where ASR is least informative: among attacks that succeed for genuinely different reasons. Across four aligned LLMs and three jailbreak paradigms, attacks with nearly identical ASR land at clearly different points on the plane: the same model can fail through different temporal patterns. The geometry matches refusal-direction probes from hidden states on most conditions, with one model showing the limit of our fixed-lexicon approach. A simple early-stop rule derived from TLO cuts successful jailbreaks by more than half, without false alarms on plain benign queries. Safety evaluation should report when and how a failure unfolds, not only whether it occurred. TLO makes the first two observable from logits alone.
Uncovering Logit Suppression Vulnerabilities in LLM Safety Alignment
Huazhong University of Science and Technology(华中科技大学);Nanyang Technological University(南洋理工大学);University of New South Wales(新南威尔士大学);Nankai University(南开大学)
LongGuard: Mechanistic Analysis and Training-Free Mitigation of Long-Context Failure in Safety Guardrails
Institute of Information Engineering, Chinese Academy of Sciences(中国科学院信息工程研究所);School of Cyber Security, University of Chinese Academy of Sciences(中国科学院大学网络空间安全学院)
ChainClaw: A Layered Agent Framework for Reliable On-Chain Execution
Beijing Advanced Innovation Center for Future Blockchain and Privacy Computing, Beihang University(北京航空航天大学未来区块链与隐私计算高精尖创新中心);Beihang University(北京航空航天大学);School of Artificial Intelligence, Beihang University(北京航空航天大学人工智能学院);School of Mathematical Sciences, Beihang University(北京航空航天大学数学科学学院);Zhongguancun Laboratory(中关村实验室)
Logit-Boundary Geometric Belief Interfaces and Sparse Sheaf-Enclave Protocols: A Self-Contained Substrate for Secure Network Electronic Health Record (EHR) Interoperability
Mitigating LLM Over-Refusal via Dynamic Semantic Routing Calibration
School of Artificial Intelligence, Jilin University(吉林大学人工智能学院);CSIRO(澳大利亚联邦科学与工业研究组织);Center of Excellence for Generative AI, King Abdullah University of Science and Technology(阿卜杜拉国王科技大学生成式人工智能卓越中心)
Action-Conditioned Risk Gating for Safety-Critical Control under Partial Observability
University of Virginia(弗吉尼亚大学);Google(谷歌);University of Maryland, College Park(马里兰大学 College Park 分校);Stanford University(斯坦福大学);Louisiana State University(路易斯安那州立大学);College of William and Mary(威廉与玛丽学院)
From Agent Traces to Trust: A Survey of Evidence Tracing and Execution Provenance in LLM Agents
Griffith University(格里菲斯大学);Jiangsu University(江苏大学);University of Southern Queensland(南方昆士兰大学);Peking University(北京大学);Great Bay University(大湾大学);Nanjing University(南京大学);Macquarie University(麦觉瑞大学);Southern University of Science and Technology(南方科学与技术大学)
ATBench: A Diverse and Realistic Agent Trajectory Benchmark for Safety Evaluation and Diagnosis
Shanghai AI Lab(上海人工智能实验室);Fudan University(复旦大学);Shanghai Jiao Tong University(上海交通大学);Tsinghua University(清华大学);KAUST(卡塔尔人工智能科学中心);East China Normal University(华东师范大学)
AI Disclosures Project, Social Science Research Council Institute for Innovation and Public Purpose, University College London(AI披露项目,社会科学理事会创新与公共目的研究所,伦敦大学学院);AI Disclosures Project, Social Science Research Council(AI披露项目,社会科学理事会);O’Reilly Media(O’Reilly媒体)