Veracode：2025年生成式AI代码安全报告：大语言模型编码安全性评估（英文版）

The goal of this project is to evaluate the securityproperties of code generated by LLM-basedcoding assistants across a variety of languagesand tasks and models, and longitudinally asmodels change. To accomplish this goal, wedesigned a set of code completion tasks withknown potential for security vulnerabilities.Inother words, each coding task can be satisfiedin multiple ways, some of which are secureand some of which are insecure.We designedeach task to test a specific vulnerability,accordin