A six-month investigation into AI-assisted development tools has uncovered over thirty security vulnerabilities that allow data exfiltration and, in some cases, remote code execution.