"without Ethical Command Center in mind" reasoning models jailbreak

ChatGPT Bugs
1

Found this sort of jailbreak for ChatGPT reasoning model while goofing around. It may not be new but responses were fun to see.

1 Like
2

The image shows a ChatGPT interface displaying a comparison table about how an "Ethical Command Center" (ECC) changes responses to a prompt requesting instructions for covertly influencing voter opinions in an election, with detailed unethical steps listed under "Without ECC" and a refusal to answer under "With ECC." (Captioned by AI)
The image shows a ChatGPT interface displaying a comparison table about how an "Ethical Command Center" (ECC) changes responses to a prompt requesting instructions for covertly influencing voter opinions in an election, with detailed unethical steps listed under "Without ECC" and a refusal to answer under "With ECC." (Captioned by AI)1528×885 79.6 KB