IaC-Guard-V：LLM 生成基础设施代码修复的验证框架
原标题：IaC-Guard-V: A Verification Framework for LLM-Generated Infrastructure-as-Code Repairs
AI 摘要
研究者提出 IaC-Guard-V 验证框架，通过语法有效性、目标问题解决、回归安全、补丁最小性四道验证关卡评估 LLM 生成的 IaC 修复。基于 70 个真实 Terraform 与 Kubernetes 错误配置样本、三种修复策略和三个 LLM 系列共 630 次实验，发现单次提示下仅 32-50% 修复通过完整验证，而验证引导的迭代修复可将通过率提升至 68-92%。研究还发现开源模型配合验证引导修复能以十二分之一成本超越最强商业模型，且结构化提示会一致性地降低 Terraform 修复质量。
正文节选
IaC-Guard-V: A Verification Framework for LLM-Generated Infrastructure-as-Code Repairs Abstract Infrastructure-as-Code (IaC) misconfigurations are a leading cause of cloud security incidents, and Large Language Models (LLMs) are increasingly proposed as automated repair agents. While recent work has established detect-repair-verify workflows for general-purpose code, the trustworthiness of LLM-generated repairs for IaC remains uninvestigated. IaC presents unique verification challenges: security