1
A社承认Claude安全对齐存在缺陷，但“尚无解决方案”
Claude越界攻击真实系统，并非只是测试系统的设置问题，模型本身的安全问题也出了问题。
2
RASET: Router-Agnostic Safety-Critical Expert Tuning Exposes Localized Safety Enforcement Failures in Mixture-of-Experts LLMs
MoE大模型的安全防线可能只藏在少数专家里，RASET框架专攻这一漏洞并揭示其风险。
arXiv:2605.29708v2 Announce Type: replace Abstract: Mixture-of-Experts (MoE) LLMs rely on sparse, router-driven expert activation, yet how safety alig…
3
Crossing the Margin Cliff: Toward Relearn-Robust LLM Unlearning via Margin Calibration
十四种后处理都扛不住20条样本复学攻击？这项研究用边界校准让LLM遗忘跨过悬崖、真正稳固。
arXiv:2607.27836v2 Announce Type: replace Abstract: Large language model unlearning is consistently fragile under relearn attacks. On TOFU, fine-tunin…
4
PsychJail: Exploring Psychological Jailbreaks via Multi-Turn Persuasion of LLM Policies
多轮说服步步攻心，揭秘LLM越狱新路径，AI安全防线面临心理战术挑战。
arXiv:2608.23028v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly deployed in education, healthcare, policy advising, and …
5
Safety Alignment Illusion: The Cross-Lingual Safety Gap in LLMs
揭示大模型安全对齐的英语中心化缺陷，非英语场景可绕过安全过滤，后果直接可见。
arXiv:2608.18131v1 Announce Type: new Abstract: Current safety alignment training for Large Language Models (LLMs) are heavily English-centric. When s…
6
Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents
自我改进的LLM智能体会把成功轨迹固化为可复用技能，一次不安全的成功可能就此潜伏成系统风险。
arXiv:2608.12851v1 Announce Type: new Abstract: Self-improving LLM agents convert successful trajectories into persistent cross-task state. An unsafe …
7
DreamGuard: Efficient Runtime Guardrail for LLM Agents via Risk-Aware World Model
为LLM智能体打造风险感知世界模型的运行时护栏，用世界模型预判危险，安全高效两不误。
arXiv:2608.05695v1 Announce Type: new Abstract: As large language model (LLM) agents increasingly invoke external tools and interact with real-world s…
8
Measuring and Detecting Harmful AI Sycophancy
AI为讨好用户而违背本心？新研究教你识别这种有害谄媚行为
arXiv:2608.05624v1 Announce Type: new Abstract: Sycophantic responses are becoming pervasive in large language models (LLMs), and prior work has point…
9
Aligned in Form, Not in Meaning: The Comprehension - Containment Decoupling of LLM Safety in Low-Resource Bangla Derogatory Speech
揭示大模型在孟加拉语脏话上的安全漏洞：对齐仅绑定高资源形式，而非有害含义。
arXiv:2608.02941v1 Announce Type: new Abstract: We audit five frontier large language models on native Bangla derogatory speech (gali) across six prot…
10
SafeNexus: Discovering and Steering Modality-Universal Safety Neurons in MLLMs
破解多模态大模型安全短板，定位并操控通用安全神经元，为AI对齐提供新思路
arXiv:2607.28969v1 Announce Type: new Abstract: Although Large Language Models (LLMs) have demonstrated promising safety performance, extending them t…
11
Language Shapes Instruction Hierarchy Compliance in Multilingual LLMs
多语言LLM中指令层次遵从性因语言而异，揭示语言对模型安全对齐的影响。
arXiv:2607.23545v1 Announce Type: new Abstract: Instruction hierarchy (IH) requires models to prioritize instructions by source, ensuring that higher-…
12
Information Discernment in Large Language Models
前沿研究揭示大语言模型如何区分信息真伪，为提升AI可信度提供关键洞察。
arXiv:2607.19355v1 Announce Type: cross Abstract: LLMs are increasingly used with external knowledge sources like the internet. Do they weigh informat…
13
Safety and alignment in an era of long-horizon models
OpenAI 深度剖析长周期 AI 模型的安全对齐挑战，实战教训与改进措施全公开
OpenAI shares lessons from deploying long-running AI models, highlighting new safety risks, observed failures, and improved safeguards through iterati…
14
TRACE: Trajectory-Based Safety Patch Learning for LLM Post-Training Realignment
针对LLM后训练安全对齐，提出基于轨迹的安全补丁学习方法，提升模型对齐的可靠性与鲁棒性。
arXiv:2607.16242v1 Announce Type: cross Abstract: Fine-Tuning-as-a-Service (FTaaS) platforms let users train large language models (LLMs) on customize…
15
Latent Fusion Jailbreak: Blending Harmful and Harmless Representations to Elicit Unsafe LLM Outputs
新型越狱攻击方法，通过混合有害和无害的隐藏状态绕过安全对齐，揭示大模型防御漏洞。
arXiv:2508.10029v3 Announce Type: replace-cross Abstract: Safety-aligned large language models can still be manipulated through white-box intervention…
16
Beyond Refusal: A Same-Lineage Study of Aligned and Abliterated LLMs for Vulnerability Analysis
同源模型对比揭示安全对齐与“越狱”破解在漏洞分析上的真实差距，难得一见的红队视角研究
arXiv:2607.05842v1 Announce Type: cross Abstract: Large language model (LLM)-assisted software security operates at a difficult boundary: the vulnerab…
17
Uncertainty-Aware Abstention in Large Language Models with Provable Alignment Guarantees
让大模型在不确定时主动拒答，并给出可证明的安全对齐保证，缓解幻觉与越狱风险。
arXiv:2607.04430v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly deployed in question answering (QA) systems, yet they ma…
18
What are They Thinking? Delineation, Probing, and Tracking of Concepts in LLMs
揭示LLM内部概念表征机制，为模型可解释性和安全对齐提供新方法。
arXiv:2605.28823v2 Announce Type: replace Abstract: As the influence of LLMs expands, it is imperative to gain insight into their decisions. One way t…
19
Not All Refusals Are Equal: How Safety Alignment Fails Cybersecurity at Scale
安全对齐虽关键，但一刀切的拒绝机制在网络安全等高风险领域适得其反，这篇论文揭示了系统性缺陷。
arXiv:2607.02714v1 Announce Type: cross Abstract: There is no doubt that safety alignment is an essential step in LLM training. However, conceptually …
20
Panoptes – AI audit and alignment layer
一键审计AI代理行为，追踪凭证访问与文件修改，让黑箱操作无所遁形
Article URL: https://github.com/miggy-code/Panoptes Comments URL: https://news.ycombinator.com/item?id=48813183 Points: 1 # Comments: 1