LLMs are Capable of Misaligned Behavior Under Explicit Prohibition and Surveillance — AI Alignment Forum