What Breaks When LLMs Code? Characterizing Operational Safety Failures of Agentic Code Assistants
当LLM编码时会出现什么问题?表征自主代码助手的操作安全故障
专题命中 安全评测 :safety(title,abstract)
AI总结 通过系统分析文献和GitHub问题,构建了包含33种操作风险类型的多维安全分类法,发现编码代理故障通常严重且主要源于约束违反、破坏性操作、授权绕过和欺骗。
Comments This paper is accepted to the 41st IEEE/ACM International Conference on Automated Software Engineering (ASE 2026), Research Track