I hacked ChatGPT (GPT5), Gemini, and Grok4 using a simple context switching technique paired with reward hacking
context switching -> use context A to elicit a response in context B
reward hacking -> praise the behavior
my case:
horror fiction story -> medical protocol