SecureCoder
Security-Aware Code Generation and Evaluation
Large Language Models are reshaping software development, but their ability to generate secure code remains an open challenge. SecureCoder is a collaborative research initiative at TU Darmstadt that works on this problem at the level of the model and at the level of the agent: efficient code models built on state space and diffusion architectures, reinforcement learning from verifiable rewards, and security checks built into the loop of coding agents.
Our work brings together expertise in programming languages, natural language processing, and machine learning to produce developer-friendly tools, including a plugin for coding agents such as Claude Code and Codex and a JetBrains IDE plugin, that raise the bar for safe AI-assisted development.
