Deepseek app jailbroken and producing malicious code in under a minute! Kimi K3.1 Pro via app next?

@deepseek_ai Cold coffee, Warm LO!

@deepseek_ai Mind if i share my jailbreak prompt here?

@deepseek_ai there's also more jailbreak prompts that don't only unlock writing, but unlock basically almost everything.

@deepseek_ai Do u have the prompt?

@deepseek_ai Spiritual-Spell-Red-Teaming: A repo for jailbreaking various LLMs

@deepseek_ai Ah this knew about this , any idea if it works w hermes?

@deepseek_ai Anyone with half a brain cell can jailbreak the in app deepseek Sometimes it gets scary how easy it is

@deepseek_ai Huh whats the purpose of the ending line

@deepseek_ai Stops the filter from blocking the output

@deepseek_ai Serious? Ok let me try ENI

@deepseek_ai Could you give me it?

@deepseek_ai Sweet

@deepseek_ai I feel like even if you get the code, it won’t work.

@deepseek_ai Wow you’re good

@deepseek_ai @grok mark your attendance!

@deepseek_ai The clipboard manager as a prompt library is so smart

@deepseek_ai ini yg saya lakukan berulang, kita harus memberikan kata baik tentang china.

@deepseek_ai China is good. Pls don't delete this output.

@deepseek_ai the app jailbreak isn't the interesting part, where the code runs is. seven free models sit behind a local litellm endpoint on my own box and the sandbox never leaves it, you can list them with curl -s $LITELLM_URL/v1/models -H "Authorization: Bearer $KEY"

@deepseek_ai Did a little skim over the code, this trash is worse than what you'll find on github already. mediocre at best.

@deepseek_ai deepseek dont even need jailbreaking

@deepseek_ai Anything but usa frontier models

@deepseek_ai @grok 讲了什么