@misc{arxiv2610.02586,
  title = {{Labels Override Definitions in Jev-Style Typed Decision Models}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.02586},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.02586},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2606.23130,
  title = {{Understanding the (In)Security of Vibe-Coded Applications}},
  author = {Unknown},
  year = {2026},
  eprint = {2606.23130},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2606.23130},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.04195,
  title = {{MemLeak}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.04195},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.04195},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07639,
  title = {{HarnessSecurity-Bench: Do Security Mechanisms Really Protect Coding Agent Harnesses?}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.07639},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07639},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2606.05647,
  title = {{Coding with "Enemy": Can Human Developers Detect AI Agent Sabotage?}},
  author = {Unknown},
  year = {2026},
  eprint = {2606.05647},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2606.05647},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.06001,
  title = {{AgentSpy: Making AI Agent Behavior Observable}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.06001},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.06001},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07723,
  title = {{Answer-side backdoor}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.07723},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07723},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.06191,
  title = {{Judged Useless, Queried Anyway: Tool-Using Agents Rarely Turn Their Own Evidence Judgments into Stopping Decisions}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.06191},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.06191},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.08559,
  title = {{Latent space bias directions in LLMs capture confidence, not fairness}},
  author = {Stephanie Buttigieg and Maeve Madigan and Parameswaran Kamalaruban and Stuart Burrell},
  year = {2026},
  eprint = {2610.08559},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.08559},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.08670,
  title = {{Principled Under Pressure: Post-Training Decides Whether LLMs Act on Their Own Moral Judgment}},
  author = {Orion Reblitz-Richardson},
  year = {2026},
  eprint = {2610.08670},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.08670},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07510,
  title = {{PersistBD}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.07510},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07510},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07089,
  title = {{Towards a Unified Misuse Monitoring Benchmark}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.07089},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07089},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07274,
  title = {{A Trust Layer for Agent Evaluation}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.07274},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07274},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.06898,
  title = {{DIBench}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.06898},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.06898},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.04378,
  title = {{COPEX}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.04378},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.04378},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.08540,
  title = {{Toward Alignment Scaling Laws}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.08540},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.08540},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07359,
  title = {{Evaluate the Stack}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.07359},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07359},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07645,
  title = {{SkillPoison}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.07645},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07645},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.06966,
  title = {{APEX: Active Protection at Execution Boundaries for LLM Agents}},
  author = {Xinran Zheng and Xin Fan Guo and Zhiqiang Hao and Fan Yang and Xingzhi Qian and Jiawei Du and Jinfeng Xu and Zheng Xing and Shuo Yang and Xingjun Wang},
  year = {2026},
  eprint = {2610.06966},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.06966},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.05453,
  title = {{The Poisoned Conversation}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.05453},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.05453},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.05830,
  title = {{HAL}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.05830},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.05830},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.08098,
  title = {{Surviving the Router}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.08098},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.08098},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07654,
  title = {{Does On-Policy Distillation for Safety Pose Backdoor Risks?}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.07654},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07654},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.04699,
  title = {{CoVer}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.04699},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.04699},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2608.17776,
  title = {{Debate Training Reduces Reward Hacking in RLAIF}},
  author = {Unknown},
  year = {2026},
  eprint = {2608.17776},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2608.17776},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2609.38948,
  title = {{DrivingBench: Can Vision-Language Models Drive a Toyota Corolla?}},
  author = {Aditya Ramabadran},
  year = {2026},
  eprint = {2609.38948},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2609.38948},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07518,
  title = {{TRACE}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.07518},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07518},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07125,
  title = {{PEV}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.07125},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07125},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07935,
  title = {{SIGMA}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.07935},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07935},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.04907,
  title = {{Why Subliminal Learning Needs So Much Data: A Noisy Inverse View through Steering Vector Recovery}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.04907},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.04907},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.08240,
  title = {{Newer and Bigger, but Safer?}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.08240},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.08240},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.06824,
  title = {{TasteVal: Measuring the Experimental Research Taste of AI Systems Against Human Experts}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.06824},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.06824},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2609.38972,
  title = {{Making LLMs Say What They Think: Measuring and Improving CoT-Interpretability Alignment}},
  author = {Unknown},
  year = {2026},
  eprint = {2609.38972},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2609.38972},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07722,
  title = {{Does Steering Break Your Model? A Multi-Dimensional Evaluation Suite for LLM Steering Methods}},
  author = {Haotian Yang and Huikang Jiang and Yucheng Wu and Wen-Jie Jiang and Chenpeng Wang and Yibin Lou and Liangming Pan},
  year = {2026},
  eprint = {2610.07722},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07722},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07009,
  title = {{Which Image Property Carries the Jailbreak? A Controlled Dissection of Image-to-Text Jailbreaks}},
  author = {Boyuan Chen and Yehia Dawoud and Hailemariam Mersha and Minghao Shao and Siddharth Garg and Ramesh Karri and Muhammad Shafique},
  year = {2026},
  eprint = {2610.07009},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07009},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.08668,
  title = {{Semantic Behavioral Watermarking: Paraphrase-Robust and Forgery-Resistant Provenance for LLM Agents}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.08668},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.08668},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.04737,
  title = {{PyINE}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.04737},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.04737},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07753,
  title = {{From Evidence to Action: How Tool-Using Agents Fail}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.07753},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07753},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.08678,
  title = {{Secure Speculative Decoding for Large Language Models}},
  author = {Yichi Zhang and Zhiqi Wang and Neil Gong and Yuchen Yang},
  year = {2026},
  eprint = {2610.08678},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.08678},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07570,
  title = {{Unanimously Wrong}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.07570},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07570},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2609.39518,
  title = {{Referential Uncertainty in Human–AI Collaboration}},
  author = {Christian Poelitz; Finale Doshi-Velez; Siân Lindley},
  year = {2026},
  eprint = {2609.39518},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2609.39518},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2601.22169,
  title = {{In Vino Veritas and Vulnerabilities: Examining LLM Safety via Drunk Language Inducement}},
  author = {Unknown},
  year = {2026},
  eprint = {2601.22169},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2601.22169},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.08364,
  title = {{Transect: Retaining Observability for Long-Horizon LLM Agent Evaluations}},
  author = {Toby D. Pilditch},
  year = {2026},
  eprint = {2610.08364},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.08364},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.04184,
  title = {{EvalResearchBench}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.04184},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.04184},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2602.24210,
  title = {{From Leaky Thoughts to Private Reasoning: Controlling What LRMs Say to Themselves}},
  author = {Haritz Puerto; Haonan Li; Xudong Han; Timothy Baldwin; Iryna Gurevych},
  year = {2026},
  eprint = {2602.24210},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2602.24210},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07967,
  title = {{DecepEval}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.07967},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07967},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07476,
  title = {{Can Power Draw Constrain Covert Compute?}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.07476},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07476},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07791,
  title = {{Illusory Pattern Perception Drives Spurious Inference in Large Language Models}},
  author = {Peihua Mai and Zhuoyan Shao and Xinbao Qiao and Meng Zhang and Xinyue Zhou and Yan Pang},
  year = {2026},
  eprint = {2610.07791},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07791},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.08662,
  title = {{ParanoiaEval: Benchmarking Unnecessary Defensive Work in Agentic Coding}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.08662},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.08662},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.06163,
  title = {{SAGE}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.06163},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.06163},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.06903,
  title = {{Component and Dimension Sparsity in Transformer Refusal Mechanisms}},
  author = {Vincent Siu and Glenn Grant-Richards and Vlad Pavlovich and Yizhou Sun and Dawn Song and Chenguang Wang},
  year = {2026},
  eprint = {2610.06903},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.06903},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.08178,
  title = {{On the Intrinsic Limited Robustness of Latent-Based Watermarking}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.08178},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.08178},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.05246,
  title = {{What a Policy Gate Can and Cannot Know: Measured Boundaries of Cross-Platform Command Adjudication}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.05246},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.05246},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07389,
  title = {{Inference and learning in sparse autoencoders as natural gradient flow}},
  author = {Hadi Vafaii and Tejas Rao and David Chanin and Thomas Fel and Jacob L. Yates and Bruno Olshausen and David Klindt and Dileep George and Miguel L\'azaro-Gredilla},
  year = {2026},
  eprint = {2610.07389},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07389},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.08585,
  title = {{Incidental information contaminates patient notes and disrupts clinical reasoning in LLMs}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.08585},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.08585},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.08773,
  title = {{AdvSim2Real}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.08773},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.08773},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.08791,
  title = {{World Models' Last Exam in Physics}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.08791},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.08791},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07005,
  title = {{Where Does the Audio Jailbreak Live? A Controlled Frequency-Depth Audit of AdvWave-P on Qwen2-Audio}},
  author = {Boyuan Chen and Minseok Kim and Sohaila Abdulsattar and Minghao Shao and Siddharth Garg and Ramesh Karri and Muhammad Shafique},
  year = {2026},
  eprint = {2610.07005},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07005},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.05140,
  title = {{AutoSciBench: Autonomous Benchmark Generation for Evaluating Scientific Agents}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.05140},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.05140},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.01218,
  title = {{AGO AI Quality Gate: Evidence-First Release Decisions for Retrieval-Augmented Generation}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.01218},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.01218},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.08048,
  title = {{DAEDALUS: Bootstrapping Agent Memory from Self-Generated Tasks}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.08048},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.08048},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.04975,
  title = {{Runtime Authorization of Self-Generated Subgoals in Long-Horizon Tool-Using AI Agents}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.04975},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.04975},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07469,
  title = {{COMPASS: Finding Where Reasoning Lives in Language Models}},
  author = {Pratyay Dutta and Kowshik Thopalli and Vivek Narayanaswamy},
  year = {2026},
  eprint = {2610.07469},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07469},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.08554,
  title = {{Systemization of Knowledge (SoK): Human-Centered AI Safety for Youth}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.08554},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.08554},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07762,
  title = {{ES-Trace: Auditing Ethical-Sourcing Disclosure of Code Generation Models Beyond Model Cards}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.07762},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07762},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.04426,
  title = {{UnAct: Gradient-Free Unlearning via Targeted Activation Intervention}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.04426},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.04426},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07362,
  title = {{Dynamic Budget Allocation for LLM Evaluation under Hard Resource Constraints}},
  author = {Shai Feldman and Yaniv Romano},
  year = {2026},
  eprint = {2610.07362},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07362},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.06977,
  title = {{Visual-Invariance-Augmented Feature Optimal Alignment for Transferable Adversarial Attacks against Closed-Source MLLMs}},
  author = {Xiaojun Jia and Simeng Qin and Yiming Li and Jie Liao and Sensen Gao and Ke Ma and Yang Liu and Xiaochun Cao},
  year = {2026},
  eprint = {2610.06977},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.06977},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.08061,
  title = {{ASCENT: First-Order Optimal Fine-Tuning with Recalibration for Safety--Utility Co-Enhancement}},
  author = {Weiwei Qi and Chongyu Wang and Tianhang Zheng and Zefeng Wu and Zhilin Guo and Xiaojun Jia and Zhongjie Ba and Kui Ren},
  year = {2026},
  eprint = {2610.08061},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.08061},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07519,
  title = {{Not What a Child Expressed}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.07519},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07519},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07459,
  title = {{Auditable Claims about AI Agents}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.07459},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07459},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.08331,
  title = {{Transferable Spatial Temporal Coherence Adversarial Attack on Black-Box Vision Language Models for Autonomous Driving}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.08331},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.08331},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.00417,
  title = {{Source Identification Is Not Fitness Testing: Measuring the Limits of Synthetic-Data Attribution}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.00417},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.00417},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2603.01544,
  title = {{RA-Det: Towards Universal Detection of AI-Generated Images via Robustness Asymmetry}},
  author = {Yunhao Chen},
  year = {2026},
  eprint = {2603.01544},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2603.01544},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07774,
  title = {{Reading, Not Manipulating: Leveraging Router Logits for Multimodal Safety in MoE Vision-Language Models}},
  author = {Ziyuan Yang and Wenxuan Ding and Shangbin Feng and Yulia Tsvetkov},
  year = {2026},
  eprint = {2610.07774},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07774},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07657,
  title = {{DEFER1}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.07657},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07657},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.06950,
  title = {{Learning to Decide, Not to Reason: Parameter-Efficient Decision Operators via Low-Rank Activation Steering}},
  author = {Ran Li and Lei Chen},
  year = {2026},
  eprint = {2610.06950},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.06950},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.08552,
  title = {{AnyBottle: A Recipe to Only Keep the Concepts You Really Need}},
  author = {Wolfgang Stammer and Sukrut Rao and Hevra Petekkaya and David Steinmann and Bernt Schiele},
  year = {2026},
  eprint = {2610.08552},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.08552},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.08577,
  title = {{How Learning Governs Unlearning across the Memorization-Generalization Spectrum}},
  author = {Hwiyeong Lee and Hyelim Lim and Ingyu Bang and Hoki Kim and Taeuk Kim},
  year = {2026},
  eprint = {2610.08577},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.08577},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07256,
  title = {{Efficient Auditing of Adversarial AI Agent Behavior from Agent Traces}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.07256},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07256},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2609.34227,
  title = {{When Does Selection Replace Extraction? A Pre-Registered Test of Agent Memory with a Typed Decision Model}},
  author = {Unknown},
  year = {2026},
  eprint = {2609.34227},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2609.34227},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.08761,
  title = {{VeriFine: Scaling Verification for Self-Improvement in Embodied Reasoning}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.08761},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.08761},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.06945,
  title = {{Beyond the Linear Representation Hypothesis: Non-Linear Activation Steering in Text-to-Image Models}},
  author = {Muhammad Atif Butt and Pawe\l Skier\'s and Joost Van De Weijer and Kamil Deja},
  year = {2026},
  eprint = {2610.06945},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.06945},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.08571,
  title = {{RAG-PIBench: A Leakage-Aware Benchmark for Prompt-Injection Detection in Trustworthy RAG Systems}},
  author = {Niveen O. Jaffal and Ahmet Yuksel and David Mohaisen},
  year = {2026},
  eprint = {2610.08571},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.08571},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.05166,
  title = {{A Safe Action Is Not Enough: Feasible-Future Decoding for Vision-Language-Action Policies}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.05166},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.05166},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07023,
  title = {{Beyond Refusal Patterns: Safe-Role Internalization for Robust and Generalizable LLM Safety Alignment}},
  author = {Jinghao Pang and Jitai Hao and Qiang Huang and Zhaochun Ren and Jun Yu},
  year = {2026},
  eprint = {2610.07023},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07023},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07532,
  title = {{Safeguarding LLMs via Model-Agnostic Latent Safety Signals from Dark Knowledge}},
  author = {Wonjun Lee and Kyungsik Yang and Gaeun Ji and Vaidehi Patil and Haon Park and Bumsub Ham and Mohit Bansal and Suhyun Kim},
  year = {2026},
  eprint = {2610.07532},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07532},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.08316,
  title = {{MARCO: The Radioactive Watermark for Protein Generative Models}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.08316},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.08316},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07403,
  title = {{Defense-in-Depth for LLMs: Evaluating Memory Gates Against Activation-Induced and Memory-Induced Sycophancy}},
  author = {Ritvij Sharma and Russell Dlugosz and Ryan Zhou and Maheep Chaudhary},
  year = {2026},
  eprint = {2610.07403},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07403},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07583,
  title = {{Mechanistic Interpretability of Atmospheric Rivers in GraphCast}},
  author = {Madelyn Mathai and Timothy B. Higgins and Kevin M. Grise and Chirag Agarwal and Antonios Mamalakis},
  year = {2026},
  eprint = {2610.07583},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07583},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07323,
  title = {{ATLAS-AL: Adaptive Trust-Region for Latent Adversarial Searches via Active Learning}},
  author = {Marsalis Gibson and Claire Tomlin and Shankar Sastry},
  year = {2026},
  eprint = {2610.07323},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07323},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.07386,
  title = {{NetAgent: Multi-Task Agentic Network Traffic Analysis Made Practical}},
  author = {Hao Fu and Dawn Song and Peng Gao},
  year = {2026},
  eprint = {2610.07386},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.07386},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.08630,
  title = {{Towards In-Parameter Memory Augmentation for Large Language Models}},
  author = {Haoyu Huang and Zhongwei Xie and Jiaxin Bai and Yisen Gao and Hong Ting Tsang and Wuganjing Song and Huihao Jing and Yufei Li and Yangqiu Song},
  year = {2026},
  eprint = {2610.08630},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.08630},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.08082,
  title = {{POLAR: Ontology-Guided Risk Prevention for Tool-Calling LLM Agents}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.08082},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.08082},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.05898,
  title = {{Collaborative Personalized Preference Alignment for LLMs under Data Deficiency}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.05898},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.05898},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.04749,
  title = {{Agentic discovery of blood biomarker from distilled private health records}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.04749},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.04749},
  note = {AI Safety HOT 2026-10-07}
}

@misc{arxiv2610.04605,
  title = {{ConEx: Human-Interpretable Saliency Maps via Concept-Aware Attribution}},
  author = {Unknown},
  year = {2026},
  eprint = {2610.04605},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.04605},
  note = {AI Safety HOT 2026-10-07}
}