Secure Code Generation at Scale with Reflexion
This addresses the issue of generating secure code for developers and organizations using LLMs, but it is incremental as it builds on existing prompting methods.
The paper tackled the problem of insecure code generation by large language models, evaluating five instruction-tuned code LLMs using a reflexion prompting approach, and found that reflexion improved average security accuracy from 70.74% to 79.43% over three rounds, with most benefits in the first one to two rounds.
Large language models (LLMs) are now widely used to draft and refactor code, but code that works is not necessarily secure. We evaluate secure code generation using the Instruct Prime, which eliminated compliance-required prompts and cue contamination, and evaluate five instruction-tuned code LLMs using a zero-shot baseline and a three-round reflexion prompting approach. Security is measured using the Insecure Code Detector (ICD), and results are reported by measuring Repair, Regression, and NetGain metrics, considering the programming language and CWE family. Our findings show that insecurity remains common at the first round: roughly 25-33% of programs are insecure at a zero-shot baseline (t0 ). Weak cryptography/config-dependent bugs are the hardest to avoid while templated ones like XSS, code injection, and hard-coded secrets are handled more reliably. Python yields the highest secure rates; C and C# are the lowest, with Java, JS, PHP, and C++ in the middle. Reflexion prompting improves security for all models, improving average accuracy from 70.74% at t0 to 79.43% at t3 , with the largest gains in the first round followed by diminishing returns. The trends with Repair, Regression, and NetGain metrics show that applying one to two rounds produces most of the benefits. A replication package is available at https://doi.org/10.5281/zenodo.17065846.