Exploiting Programmatic Behavior of LLMs: Dual-Use Through Standard Security Attacks