How We Stopped Claude Code from Writing eval() in Production

A team discovered Claude Code had generated eval() in production code that passed all reviews and tests but created a remote code execution vulnerability. After auditing three months of AI-generated code, they found shell injections, XSS vulnerabilities, and fabricated statistics that existing security tools missed. They built Quadruple Verification, an open-source plugin that blocks dangerous patterns at generation time before files are written.