SAFESHIELD: A Decision-Organization Framework for Deployment-Time Safety of Small Language Models
Deployment-time safety of language models is commonly implemented through runtime guardrails such as input moderation, r
#cs.SE
共收录 457 条相关情报。
← 返回所有主题Correct Verdicts, Flawed Reasoning: Structured Auditing of LLM-based Vulnerability Reasoning
Large Language Models (LLMs) are increasingly deployed for automated software vulnerability analysis. Binary classificat
Where Did the Repair First Go Wrong? Localizing the Origins of Silent Failures in Agentic Vulnerability Repair
Localizing where an LLM-based agent first fails to uphold security during a repair can show which stage of its workflow
Security Is More Than a Library Call: How Security Features Live in Code
Implementing security features---functionalities that protect sensitive data or prevent malicious actions by attackers--
Backdooring Sparse Autoencoders
Sparse autoencoders (SAEs) are increasingly used not only to interpret language models but also to intervene on their in
VulValidate: Auditing Function-Level Vulnerability Labels with Executable Evidence
基于学习的漏洞检测高度依赖训练标签的质量。现有大规模数据集（BigVul、PrimeVul、DiverseVul 等）通常以「漏洞修复提交」为依据，把被补丁改动过的函数直接标注为「存在漏洞」，但被安全补丁修改并不等于该函数本身存在漏洞，由此
Characterizing Security Effects of OSS Vulnerabilities in Agent Systems
Software agents increasingly depend on open-source components when executing tools and interacting with external systems
When Talk Isn't Code: Comparing LLM Agents That Simulate and Develop Software
LLMs are used both to simulate network protocol implementations, as honeypots do, and to write them. Prior work evaluate
PrivDev: Mapping Static-Analysis Data Types to DPV
Static-analysis scanners can identify personal-data types in source code, but they lack mechanisms to connect these find
Safety Must Survive Self-Improvement: Why Failures Persist and How Agents Recover
该论文研究递归自我改进（RSI）场景下智能体（agent）的安全性问题：当 LLM 编辑器跨代际持续优化可执行的智能体组件时，如何既阻止不安全行为在后续版本中延续，又能在失败发生后实现恢复。作者构建了一个受控测试台，使用有状态的授权（aut
ABSENTIA: Detecting Broken Access Control Vulnerabilities in Web Applications
该论文聚焦 Web 应用中最为普遍的安全风险之一——失效的访问控制（Broken Access Control, BAC），即授权机制的缺失或错误。作者指出，与注入类漏洞本质不同：注入是“不可信数据流入危险操作”的数据流问题，而授权是一种“
Aletheia: Permission-Minimality Testing for Coding-Agent Rules
该论文关注编码智能体（coding agent）所依赖的「仓库指令文件」（如 AGENTS.md、copilot-instructions 一类指导性规则）带来的安全风险：这些文件本意是引导智能体正确完成开发任务，但也构成一条 prompt
Approval Laundering: Systematizing Approval--Execution Binding Failures in AI Coding-Agent Harnesses
论文聚焦 AI 编码代理 harness（Claude Code、Codex CLI、Cursor 等）的一个未被充分检验的安全假设：人类批准的动作 A，与 harness 实际执行的动作 A' 是同一个动作，其中 A 由作用域授权或会话级
A Function-level Dataset of Vulnerable and Fixed Source Code in JavaScript and TypeScript
该论文针对 JavaScript 与 TypeScript 生态中自动化漏洞检测长期受限于高质量训练数据不足的问题，构建并公开发布了函数级漏洞数据集 JsVul。与常见的多语言通用数据集不同，JsVul 采用面向 JS/TS 的语言特定处理
The Compiler May Read It, the Agent May Not: Keeping Part of a Research Code Away from a Coding Agent
这篇论文（cs.AI / cs.SE）聚焦一个在「AI 编码智能体辅助科研开发」场景中日益尖锐但缺乏系统分析的矛盾：同一份代码仓库里，某些模块是构建物理求解器所必需的（编译器必须读取），但同时又属于不宜暴露给外部编码智能体的知识产权（age
LLM-Assisted Automatic Security Proofs for Cryptographic Protocols: How Far Are We?
本文针对一个此前缺乏系统认识的问题展开研究：大语言模型（LLM）在密码协议的符号化验证（symbolic protocol verification）中究竟能起到多大作用。作者指出，LLM 已在软件工程与安全分析任务中表现出较强潜力，但其在
Trajectory-Level Security Debt in LLM Coding Agents
该论文关注LLM编码智能体（coding agent）在解决任务过程中产生的安全债务问题。作者指出，编码智能体在提交最终方案前会经过数百个中间代码状态，而现有评测通常只看最终产物，导致安全发现（security findings）随轨迹演化
When Valid Tool Calls Change Meaning: Formation-Consistent Dispatch for LLM Agents
该论文聚焦于工具增强型 LLM 智能体(tool-enabled agents)在调用分派环节的一类结构性安全缺陷。作者指出：智能体在生成工具调用时，依据的是模型可见的接口描述(descriptor，即 schema、名称、参数说明)，而宿
SmartMemory: Detecting On-chain-off-chain Communication Inconsistency for Smart Contract via Memory-based Agent
该论文研究智能合约中链上-链下通信（On-chain/Off-chain Communication, OFC）引发的安全问题。随着跨链桥、现实世界资产代币化（RWA）以及法币背书稳定币等应用快速增长，OFC 已成为去中心化金融的关键基础设
Faultless: A Program Equivalence Technique for Validating and Evaluating Neural Decompilers
该论文聚焦神经反编译器（neural decompiler）的可信度问题。神经反编译器是采用机器学习模型完成反编译任务的系统，作用是把低层语言（如汇编/中间表示）提升为高层语言（如 C 代码）。相比传统确定性反编译器，它的核心优势在于能够以