The 2025 GenAI Code Security Report examines how newer large language models perform when generating secure code across multiple programming languages and vulnerability categories. As AI models evolve rapidly, Veracode’s updated analysis evaluates more than 100 LLMs to measure their security quality, track changes over time, and compare performance by model size and CWE type. The findings show ongoing variation in security outcomes, highlighting both improvements and persistent weaknesses. This evolving landscape underscores the need for continuous evaluation and stronger safeguards when relying on AI-generated code.

Join for free to read