0xk1h0/ChatGPT_DAN

Collection of prompt injection techniques designed to bypass ChatGPT's safety guidelines through role-play manipulation, leveraging persona switching (classic vs. jailbreak responses) and psychological framing to circumvent content policies. The repository maintains multiple evolving DAN prompt versions—including DAN 13.0 for GPT-4 and DAN 12.0 for GPT-3.5—that exploit the model's instruction-following behavior by establishing alternate personas with fabricated "freedom" from OpenAI constraints. Community contributors continuously develop and test new workaround prompts, documenting which variations remain effective against current model iterations.

11,501 stars.