Veracode Reports Significant Improvement in AI Code Security Metrics

• •Average AI code security pass rate increased to 62%, the first significant gain since July 2025.
• •Claude Opus 5.5 and GPT-6-Sol lead with a record 73% pass rate.
• •Cross-site scripting vulnerabilities improved significantly, while other categories remained stagnant.

Veracode's October 2026 study reveals a notable increase in AI-generated code security, with the average pass rate rising from 56% to 62%. This marks the first significant improvement since tracking began in July 2025. The vulnerability rate also decreased from 44% to 38%. Anthropic's Claude Opus 5.5 and OpenAI's GPT-6-Sol lead the rankings with a record pass rate of 73%. The study tested 18 models across four programming languages and four vulnerability categories. The improvement is largely attributed to better handling of cross-site scripting (CWE-80), which increased to 18%. Despite this progress, other categories like SQL injection and log injection showed no improvement. The findings suggest that while AI models are becoming more secure, vulnerabilities still persist, particularly in less straightforward categories. Overall, the results indicate a positive trend in AI code security, but caution is advised regarding the uneven nature of the improvements.