Veracode’s CTO Jens Wessling discusses the 2025 GenAI Code Security Report which reveals major differences in LLM code security. Testing 20 leading models, OpenAI’s GPT-5 Mini and GPT-5 achieved the highest secure-code scores, outperforming Claude Sonnet 4.5 and Gemini 2.5 Pro. The findings highlight new risks for AI-driven development