Repository context shapes security of code from large language models
A.S.E 2.0: Understanding How Repository Context Impacts the Security of LLM-Generated Code
Cryptography and SecurityArtificial IntelligenceSoftware Engineering
Summary
When AI tools generate code for software projects, the surrounding code they see influences how secure the new code is. The authors found that AI helpers doing well on making code work doesn't always mean the code is safe from hacks. They created a benchmark of real security problems to test how different ways of showing context to AI impact security. Their findings show that missing security checks in the context or showing insecure examples can make AI write vulnerable code more often.
What this means in practice
- •For software security teams: Improve security reviews by selecting and presenting repository context that highlights safeguards when using AI code generation tools.
- •For devops teams: Configure AI coding assistants to include security-related code context to reduce the chance of generating exploitable code in deployments.
Authors
Jiongchi Yu, Keke Lian, Junqing Wang, Zhikang Fan, Zhaoxuan Li, Yongheng Huang, Ziming Zhao, Dong Zhang
Abstract
Large language model (LLM) coding assistants and agents construct a repository context before generating code. For security-sensitive code, this context decides whether the model sees the project's safeguards (e.g., authorization checks) or insecure code to imitate. However, context construction is designed almost exclusively for functional correctness, and secure-code benchmarks treat the context as fixed, so how repository context affects the security of generated code remains unclear. We construct CASE-Bench, a benchmark of 145 repository-level tasks from real CVEs in 51 projects. Each task runs in Docker, has functional tests and a proof-of-concept exploit validated in both directions, and is annotated with the repository code that enforces or undermines its security. Using CASE-Bench, we compare 13 context construction methods and six controls with six LLMs, edit single context elements, and analyze coding-agent trajectories. We find that (1) functionality-oriented context raises the functional pass rate by up to 8.2 percentage points (pp) but the secure pass rate by at most 3.7 pp, (2) withholding a safeguard lowers the secure pass rate by up to 8.1 pp, while one inserted insecure exemplar raises correct-but-exploitable code by 6.3 pp, and (3) coding agents observe the relevant safeguard in only 27-36% of trajectories. These findings inform security-aware context construction and secure-code evaluation.