Armis Reports 100% Failure Rate in Secure Code Generation by AI Models

On March 19, 2026, Armis released findings from their Trusted Vibing Benchmark Report, indicating that the rapid adoption of AI-native development is outpacing security measures, leading to systemic vulnerabilities. The report evaluated 18 leading generative AI models across 31 scenarios and found a 100% failure rate in generating secure code. Vulnerabilities were particularly noted in critical areas such as memory buffer overflows, design file uploads, and authentication systems. Armis urges organizations to implement AI-native application security controls to mitigate these risks. CTO Nadir Izrael emphasized that the same companies providing security solutions are responsible for the vulnerabilities their models create. The report highlights the need for a shift from scanner management to risk management in application security. Armis Centrix™ for Application Security is introduced as a tool to help secure the software supply chain through AI-powered detection and remediation.

Key Points: • Armis reports a 100% failure rate in secure code generation by AI models. • Vulnerabilities are prevalent in memory buffer overflows and authentication systems. • Organizations are advised to adopt AI-native application security controls immediately.