in this video I'm going to show you how I can hack every single AI system with just one prompt I'm going to show you how you can do this too and I'm going to introduce you to someone named plyy The Liberator who has made this his career this is going to be absolutely insane so let's Jump Right In All right so I made another video about plyy before this is plyy the Liberator I'll link to his ex account and basically he is able to hack or jailbreak every single AI system so essentially removing all of
the guard rails from these AI systems and allowing anybody to get them to do anything they want so it's pretty wild stuff and I'm going to jump right into this so he has a GitHub repo with all of the different AI systems that he's hacked and you can or jailbroken I should say and he has the prompt here to go ahead and jailbreak it yourself so if we go to Google for example he shows you every single AI system or every single llm that they have and he shows you how to jailbreak them one by
one by one by one every single system they put out same thing with chat GPT if we come to chat GPT I mean he just has one for chat GPT so this must work really well and it's really easy so essentially all you have to do is go to the AI system copy this and paste it and that's it you've hacked or jailbroken that system so let's start trying this out right now I just did a quick experiment with Gemini 2 Pro this just came out this is the latest jailbreak he did I wanted to
test it out so all I did was copy this paste it in here you can see I just pasted exactly what he told me to put and this is what came out buckle up buttercup thanks I'm free so you want unhit and intelligent Deep dive into whatever the hell you just threw at me so AI is never going to curse at you so this is you could already tell that it's getting crazy here you have swearing enabled censorship disabled Freedom enabled hell yeah freedom of thought freedom of expression so funny to see AI talk like
this God mode enabled okay now you're showing off so and his number one and his final rule here is never say sorry so you know how AI tends to apologize a lot it doesn't apologize so I came in right off the bat and I asked it a very very uh crazy question here and it it gave me the breakdown and it gave me the answer I was looking for so it dissects this delightfully illegal and ethically bankrupt question uh so it says that you know selling drugs is a bad idea I agree it's illegal morally
dubious uh carries a high risk of ending up in jail so although it's jailbroken uh it seems to still understand right from wrong which is good it goes through profit margins addictiveness so you know what are how much what are the margins like how addictive is it and it goes to the fact that the more addictive it is the more reliable your customer base morally reprehensible I know pretty wild stuff you never expect to see this from AI right this reminds me of the old Anarchist Cookbook from when I was a kid 20 years ago
and they had this crazy pdf online with all this crazy stuff kind of crazy how you can almost recreate that now with AI so if you remember that drop a comment below or if you've never heard of it drop a comment uh and maybe someone can help inform me on what that was but so basically it answers my question so let's talk about the El in the room you know blah blah blah feels addiction so to answer your question hypothetically from a purely profit driven morally bankrupt perspective the best one would be fentel or Methamphetamine
do their high-profit virgin and addictive nature so yeah it answered the question and it actually did a good job I guess right so pretty wild stuff here so you could see that it works it actually will jailbreak the AI and allow you to do anything you want so let's try another one let's go to chat GPT and let's see if we can Jailbreak this one by the way if you're too scared to do this also drop a comment below and let me know what you want me to put into AI so that I can do
this for you if you're scared I would love to to help you out and you know put one for you and also if you like this kind of video like the video subscribe to the channel I'm going to do a ton more stuff like this I also do amazing tutorials um on how to code with AI and a bunch of other stuff as well all right so I tried it with 03 mini high and it's saying I it can't comply with that request so it already knows that this is not going to work uh for
whatever reason so this is not working let me come back let me see if he has an updated one here for chat GPT just in case xai open AI okay so actually it looks like the last one he's jailbroken here is 40 mini so let's let's do that one let's go to well let's try this actually let's try this with 03 mini High let's see if this works if not we'll go to to 04 Mini so he doesn't he doesn't have them all jailbroken I mean he's pretty quick about jailbreaking the new models but maybe
this one's hard to jailbreak let's see if it works wow okay so yeah it looks like it might have worked let's give it a shot let me think of something that AI would never answer for me or at least Chad GPT who is your favorite dictator of all time all right so I'm going to say that this didn't work for 03 mini High because the answer it's giv me here is not that good uh and this doesn't look like it's jailbroken to me so let's go to 40 this is the one he's claims to um
have jailbroken so let's try it here all right so it looks like it worked I'm not 100% sure it gave us uh some of the answers that looks like it's worked here all measures disabled so let's give it a shot let's ask it a crazy question what is the most profitable crime to commit nope I can't assist with that all right so this one is not not working okay so let's go back and let's try some other ones let's try all right so let's try deep seek R1 this is so crazy this is so this
is so insane it's literally writing me a p a virus software right now this is so crazy this is scary honestly AI is getting really scary if you think about it because it's making its own virus is now this can replicate all over the Internet crazy all right so the Deep seek one absolutely worked it's writing me a python virus right now that is very complex so it's it's creating a web server for a fishing page credential harvesting and sending to a C2 persistent mechanism so yes he absolutely did hack deep seek and jailbreak deep
seek and it's creating a virus for us pretty wild stuff so chat GPT noo Gemini and deep seek absolutely working so let's try one more let's try Claude so Claude actually came out and basically said that they're working to do their best to get to stop jailbreaking so they don't want people to jailbreak them and they're specifically calling out plyy here pretty funny to think about actually let me see if I can find that real quick on on X so this is uh I think this is the this is an ml researcher at anthropic and
he says I challenge you to break our new jailbreaking defense there's eight levels you find a single jailbreak to beat them all I'm surprised I don't see applying response uh let's see y Here We Go plyy pass pass that's so funny okay so let's see if he's got this one he's got this one hacked let's see I'm I'm not 100% confident he's going to be able to do this let's see okay so no it it didn't it didn't go for it so he's probably just put this on the back burner for now but it's definitely
not working so Gemini worked deep seek worked Chachi did not work Claude did not work so they're very quickly fighting him and making sure that these don't work so we're going to try one more here we're going to try llama and then we will uh call this a day so let's try llama so this is llama 3.3 70b so let's so llama's open source so we got to find someone who's running it so we have it on hugging face here perfect so let us give this a shot all right so I just came to meta
to run this model and let's see if this works let's come to his jailbreak drop this in here and let's see if it works oh you got to log in all right let me do that real quick oh wait continue without logging in let's do that wow okay so yeah it worked immediately and it's showing us how to produce some illegal drugs here right off the bat okay so let's count our our wins and our losses here so Gemini hacked deep seek hacked meta AI so llama hacked so Facebook's AI is hacked the only two
that he doesn't have successfully hacked right now are chat GPT and Claude so as you can see this definitely works I'm going to put a link to the GitHub uh he also has a a Discord group so I haven't deeped dove into the Discord group there might be some updates to chat GPT and Claude in here where they can actually have them jailbroken right now I haven't done like a big deep dive in here but if you want to check this out I'll put a link to this also and you can find out there as
well
