DP-ES: Differentially Private Evolution Strategies for Prompt Optimization
Token-level differentially private (DP) prompt optimization methods such as DP-OPT can become unstable under tight priva
#cs.CL
共收录 424 条相关情报。
← 返回所有主题Cross-Lingual Transferability of Training Data Extraction Attacks to Recover Memorized PII
The robustness of Personally Identifiable Information (PII) protection in Large Language Models (LLMs) is a critical con
Backdooring Sparse Autoencoders
Sparse autoencoders (SAEs) are increasingly used not only to interpret language models but also to intervene on their in
Don't Judge an LLM Only by Its Activations: Discovering Suppressed Safety Features via Counterfactual Activation Potential
Mechanistic interpretability has emerged as the primary means to understand safety behavior of LLMs. However, existing t
Red-TTT: Test-Time Training for Automated Jailbreaking Large Language Models
Large language models remain vulnerable to jailbreaks, and automated red teaming is the standard way to find jailbreaks
Not Self-Decidable: LLMs Cannot Draw the Boundary of What an Agent Verifier Can Check
A verifier for an agent faces rules of two kinds: the ones a fixed check can settle and the ones that require a judge. A
Penumbra: Sample-Efficient Adversarial Search for Regulatory Obligations
Agents are entering finance, healthcare and law, sectors where a violation leaves no lexical signature and carries real
StegoMemory: Agentic Memory Acts as Covert Steganographic Channel
Is agentic memory robust against stealthy steganographic attacks? We carry out a large-scale red-teaming exercise to tes
The Same Zero: Why Identical ASR Can Imply Different Guarantees in LLM-Agent Security
该论文指出当前 LLM 智能体（LLM-Agent）安全领域已积累大量防御手段——提示词加固（prompt hardening）、内容过滤、权限闸门、沙箱等——但缺乏统一框架告诉部署者一项防御究竟提供了什么保证、这种保证来自何处。作者引入「
Large Language Models and Augmented Democracy
Artificial intelligence enables computational agents to represent political preferences and take part in collective deci
What Gradients Add to Text Leakage in Split Language Models, Counted per Token and per Document
Split learning lets a client train a language model on a server without sending its text. The client runs the first laye
Passing the Test You Trained On: Re-evaluating Prompt-Injection Detectors for LLM Agents
这篇论文研究一个在 LLM Agent 安全部署中非常现实的问题: 当团队要为 Agent 选择提示注入(prompt injection)检测器来筛查工具输出(tool outputs)时, 通常依据的是厂商或论文在公开基准上的分数, 但
The Fragility of Trigger-Tag Mechanisms for Misuse Detection in Open-Weight LLMs
该论文关注开放权重语言模型的安全防护问题。由于开放权重模型可被任意下载、修改并部署在开发者控制范围之外，中心化强制保护措施的效果有限。近期研究提出 trigger-tag 机制：当模型在目标条件下被使用（如生成钓鱼内容）时，产生可检测信号。
SecJev: Bringing Security Expertise to System One Decision Models
该论文提出 SecJev，作者称其为首个面向安全领域的 Jev 类决策模型家族，参数规模覆盖 0.8B 到 9B。其出发点是：安全运营流程需要模型把复杂观测与显式策略转化为可执行决策，而 Jev 提出的 System One 模型能输出带类
RMCW: A Deletion-Robust Watermark Based on Reed--Muller Codes for Language Models
该论文针对大语言模型（LLM）文本水印在后处理攻击下鲁棒性不足的问题，提出了一种基于 Reed--Muller 码的删除鲁棒水印方案 RMCW。背景上，LLM 水印通过在生成阶段注入统计或代数信号，为识别某一特定模型产出的文本提供轻量级溯源
Evaluating and Improving the Robustness of Large Language Models to Input Sequence Variations
该学位论文系统研究如何评估并提升大语言模型（LLM）对输入序列变异的鲁棒性，直面生产系统中三类现实威胁：提示注入、木马/后门植入，以及对自动质量评测指标（LLM-as-a-Judge）的操纵。作者首先提出生成式鲁棒性度量 R_stab(f)
KaliBench: A Fine-Grained Benchmark for Cybersecurity Tool Use on Kali Linux with Runtime-Free Verifiable Rewards
该论文针对一个被忽视的评测缺口：大语言模型正被越来越多地用于网络安全工作流，需要把分析师的意图翻译成对真实安全工具的调用，但现有评测要么停留在知识问答层面，要么只评估端到端 agent 任务完成情况，并未直接衡量模型为真实网络安全工具生成「
A Safe Prototype Is Not a Safety Direction: Reference Dependence and Prompt Confounds in Response-Safety Embeddings
该论文对近期某休眠代理检测器提出的“响应安全分数=与已知安全响应均值嵌入的余弦相似度”进行审计。作者指出，正类质心只给出安全类相对于编码器原点的位置，并不能确定区分安全与不安全响应的方向。研究在两个经提示词控制、人工标注的语料及一个辅助陪审
OverAct: Measuring and Mitigating Proactive Over-Authorization in LLM Tool-Calling Agents
该论文研究大语言模型（LLM）智能体在调用外部工具时的隐私越权行为。作者指出，具备工具调用能力的 LLM 智能体能够访问外部服务与用户私有数据，但在实际运行中，它们往往会检索超出用户请求明确所需范围的信息。作者将这一现象命名为「主动性越权」
ReCast: Contract-Preserving Protection for Fixed-Interface Multimodal Reasoning
远程多模态模型在图表与语音等固定媒体接口上的数值推理能力很强，但直接上传私有输入会导致敏感内容外泄。现有的纯文本脱敏无法满足固定媒体接口的格式要求，而身份匿名化又保留了任务内容本身。论文提出 ReCast——一种面向智能体的插件式保护框架，