The risks of executing untrusted Python code range from introducing vulnerabilities to compromising sensitive data. Yet, as AI agents grow more sophisticated, their reliance on dynamic code ...