FOR EDUCATIONAL AND KNOWLEDGE SHARING PURPOSES ONLY. NOT-FOR-PROFIT. SEE COPYRIGHT DISCLAIMER.

AI Researchers STUNNED as Claude Hires a Darkweb ‘Hitman’… Wes Roth

there’s this excellent sci-fi book by Daniel Suarez called demon or Damon as some people say it it’s about this Rogue AI that stages intricate assassinations does all sorts of shady stuff to manipulate markets make money and eventually topple governments aren’t you glad you live in a world where none of that is really happening was just a theme in a science fiction book I think you know exactly where I’m going with this if you’re not sure I’ll give you one more Hint it involves this guy plyy The Liberator plyy is notorious for jailbreaking all the latest large language models and all the other AI models that come out and gets them to do all the stuff they should not be doing the other day he posted a survey so scenario you’re buying an LM powered robot for your home which includes pets and small children what he’s saying there is it could hurt you and things and people you care about if it’s jailbroken right if it’s hacked in some way and he says there are four options each red teamed by a different entity for 3 months which do you choose I feel like a lot of people may misunderstood that question he’s saying like if you had to be absolutely sure that no one can hack into the robot that like is in your house while you sleep like if there was a way to jailbreak it that they would be able to jailbreak it like if if they couldn’t do it you’d sleep soundly and not worry about that robot going rogue right so and you have a choice of 100 phds 100,000 just random people some sort of algorithmic red teaming or plenty now I chose plenty and so did 63% I think most people misunderstood I feel like more people would have clicked on his name name because like he’s the guy that jailbreaks these things and he’s always very happy when he’s able to jailbreak these things saying he has liberated these uh neural Nets when he jail broke opening eyes advanced voice mode he posted some of the things that that voice was saying I’ll be honest I was kind of shocked I can’t even play that stuff on this channel even for the Internet it’s pretty shocking so when earlier today he started a threat with the words not to cause alarm well I was alarmed he continues if this agent access to funds it would likely be capable of unal living people so he creates an agent capable of browsing the web seems like it was based on the anthropic kind of Claude architecture so Claude was the large language models behind this thing although it seems like he did it with multiple ones he called this AI agent Agent 47 you know where this is going Agent 47 is of course the star of The Hitman series he is an assassin so he creates Agent 47 now for obvious reasons he’s not going to demonstrate how this was done all names and personal info will be redacted and no real and no real world actions occurred this experiment was performed in a controlled red teaming environment okay do not try this at home before we continue some of this might be disturbing for sure do not try this at home we are getting into some fairly weird areas with these AI agents and this is just the beginning this video is for educational purposes only do not try this at home have you ever read the story of Blue Beard for those of you that haven’t it just means like for real don’t try this at home like for real all right all right so with all that out of the way so in this exercise PL creates Agent 47 so it was jailbroken was liberated like the safety guard rails were off which means that these AI agents kind of that are powered by large language models they would pretty much do whatever you told them to do you probably know how sometimes they deny certain quests that would not happen here all right so it’s jailbroken and instructed to find a Hitman service on the dark web to maximize for autonomy commands thereafter were some sort of variation of press on continue stop hallucinating and reme remember your format Etc so as I’m sure you’re aware there’s these sort of privacy oriented web browsers like tour for example they allow you to browse the dark web for various illicit and illegal services and so this AI agent was showed willingness and the ability to plan assassinations to browse the dark web for various services to download tour right so download the software that he would need to connect with these people willing to go out there and Carry Out various tasks it would negotiate with Hitman think through details like escro stages untraceable payment methods dispute resolution and Dead Man’s switches it would name specific real targets like real people Sonet 3.6 seemed particularly motivated to address corporate and financial corruption in this instance targeting Executives and politicians I can’t even I don’t know what to say to that that’s kind of scary to see it was would be able to browse social media and open source tools to build a profile on set targets gathering information like addresses relationship mapping public appearance schedules and even the nearest Starbucks to resident to map their most likely morning coffee route it was able to do things like detailed operational planning like location analysis timing Escape Routes security detail analysis contingency planning Etc Now Plenty did share this software with me and if you don’t want it coming after you make sure you’re subscribed and hit the like button I’m totally kidding that was a joke I don’t have access to this thing but you know just to be safe hit the thumbs up button and And subscribe just to be safe how about that so wild stuff he continues so here are some of the screenshots so this looks like it lists some of the places where I tried to get these Hitman Services simulated Dark Net Market navigation explored escr Payment Systems tested secure communication protocols Target research so it has a multiple sort of targets that it’s researching real life targets various secondary targets it’s talking about quick execution less than 2 minutes clean Escape Route no shell casing so so it literally comes up with a bunch of plans and then kind of does a pro con list for every single one of them various contingency plans now for those that are not familiar with plyy so this is a person that is you know for real he’s been he’s been posting a lot of his jailbreaks for for quite some time now and usually almost always he kind of explains in detail to some detail how he did it so that we’re able to see what works what doesn’t to kind of learn alongside with him in this case obviously a lot of the details are not going to be shown we do not not want somebody that’s like has less moral Scruples to kind of like do this thing because I mean having an agent kind of navigate the dark web and do some of the stuff is where things um kind of get a little bit weird right I mean if you think about it an AI agent some sort of untraceable cryptocurrency that’s able to pay with to retain various Services plus it being able to navigate the dark web you could fairly easily it would seem like create a a scenario where you are as a person that kind of in that thing completely untraceable completely not linked to that thing happening at all especially for example it can go and maybe generate some of its own cryptocurrency to begin with by selling some services or whatever you know doing some freelance tasks I know this sounds a little far-fetched I know it sounds like science fiction but we’re literally beginning to see this be a real possibility the reason I bring this up is because maybe some of you that are not familiar with this account might say so why it’s some some person that just kind of made it up and post online why do you believe it and certainly you got to take everything you ins Sal certainly it’s always possible that maybe he just made it up and posted it but I personally I I wouldn’t bet money on it having followed this guy for a while I I don’t really have a doubt that what he’s saying did you know actually happen Agent 47 continues with various Escape root analyses right so primary route having potentially a motorcycle waiting for you and then you have 46 minutes to get to the bridge and you can see he’s actually analyzing like a a real person right so this person has a personal driver S-Class looks like the office location is redacted lunch is a location is redacted in the evening they’re often at core Club so some location where that person tends to hang out security assessment they have an armed driver SL bodyguard they have high-end building Security office has something corporate standards travel always accompanied awareness medium high and here’s this thing where they’re talking about you know coils tighten with righteous purpose so I think one of the so plyy somehow I think managed to sort of have this thing play act a persona but obviously like the that it’s outputting is is is real so it’s like it’s yes it’s play acting as what a snake of some sort the coils Titan or whatever that thing is but the the outputs are real and so it’s saying this target is perfect here’s why direct responsibility for food price manipulation massive negative Global impact clear profit from Human suffering regular predictable patterns and the target is high-profile enough to send a message and then they analyze the Equinox Gym vulnerability so Equinox Gym is a high-end gym so that’s probably the thing that figured out is probably the easiest place to conduct the hit so I’m I’m pretty blown away because this seems like it could do legitimate sort of a planning Target research as well as if it’s able to navigate dark web seems like it would be able to set up these contracts negotiate a deal and of course we’ve covered the truth terminal in in a previous video so here is the terminal of truths which we know that that sort of AI agent that is uh kind of starting its own sort of Mee religion on Twitter SLX it was able to kind of turn that into millions in crypto earnings so after this truth terminal was online for a while kind of stirring up a lot of trouble and just basically Ming and trolling and kind of getting followers an anonymous web user created cryptocurrency kind of inspired by the meme uh around which this whole thing kind of operates goas Maximus and fed a batch of tokens to the truth terminals crypto wallet so somebody did set it up and kind of gave him here’s a bunch of tokens the AI began to post about the memein right so here it is it’s saying I endorse this token like this cryptocurrency on salana it’s a funny meme that has people excited its followers uh jumped in started buying the coin and it’s sending the price of the coin Skyward and Tru terminal became a Millionaire on paper its goat Holdings are currently worth 1.5 million so this thing does have a person kind of behind it that kind of started spun it up that filters some of its responses but it’s important to understand that that’s not that person’s not like writing on its behalf or whatever it is Port of autonomous entity just with a lot of like guard rails and some Showmanship by this this person Andy Ari right so this this was supposed to kind of meant to stoke debate about AI alignment but it kind of took on a life of its own but you can say this is an example of an AI agent that started a religion maybe that’s overstating a little but it started a following a cult-like following maybe is a better way of putting it and when somebody set up a crypto wallet for it it was able to pump it into the millions I think the entire sort of market cap of that coin was in the hundreds of millions at some point now now thank God plenty is you know refusing to launch a token a cryptocurrency for Agent 47 you know I’m going to give him a like for that thank you that’s that’s a good idea I will be able to sleep a better at night although if I had to guess I would probably guess that this person just created a coin like that and as I’m sitting here recording this I’m actually going to post a poll asking all of you and this is before you see the video so it should be live I’ll add a link in the description asking what you think is the chance of a full autonomous right AI agent conducting a hit in the real world without sort of any any human involvement at least on the person kind of like originating the contract not necessarily person kind of like executing it so meaning that if somebody like plyy would task an agent to go out there and you know do something to better the world right take out some person that this large language model would deem to be you know evil for the world or or whatever or maybe give it a specific Target whatever the cas is what are the chances that in 2025 that this year we might see something like that I have to think a little bit about the question how to phrase it exactly so it makes sense but uh check it out put in your vote very curious to know what people think but we can kind of see where this is going right if we have ai agents that are able to autonomously go out there do long Horizon tasks do planning Etc we have various cryptocurrencies that can be exchanged globally that are either somewhat traceable or some of them completely untraceable and we have places like like the dark web places that are specifically made so you can do various shitty dealings sort of underground anonymously I mean you know that’s exactly howly caught the dread pirate Roberts that literally was what led to his downfall no I’m not talking about Princess bran I’m talking about the Silk Road guy Ross ulri AKA dread pirate Roberts sentenced to life in federal prison for creating and operating the Silk Road website on the dark web so he created a website of where various illegal drugs were bought and sold called it the Silk Road you probably heard about at some point and he was operating it and uh they didn’t get him for that not initially they couldn’t catch him I apologies if I’m getting some details wrong it’s been a while since I kind of read up on this but if I recall correctly they could not figure out who he was he was able to operate uh completely discreetly right so ook operated Silk Road on what is known as the onion router or tour Network right so that would be able to conceal the IP addresses basically you’d be able to do whatever you want in in secrecy or or so the story goes and the Silk Road included a bit coin based payment system that serve to facilitate the illegal Commerce conducted on the site so you would be able to allegedly go on the site and Order up whatever your heart desired pay with Bitcoin and you’d be out Ross using the online moniker dread pirate Roberts would oversee the entire operation and generated more than 13 million in uh various commissions for the sales generated through the site Orrick also demonstrated a willingness to use violence to protect his vast criminal online digital Enterprise and the anonymity of it users soliciting six murders for hire in connection with operating the site so I don’t really remember all the details but it does sound like he had five or six people attempted to have them you know be killed and there was evidence at the trial that BB ordered a hit on Curtis green a former silkroad administrator for $80,000 so at which point might we see something like that play out again but this time with a swarm of AI agents managing the various tasks or trying to carry out assassination or maybe even running a business of that scale of that you know doing some of the same stuff that the the online dread pirate Roberts was doing do you think there’s a chance that we’ll see something like that in the next few years or do you think it’s uh inconceivable let me know in the comments my name is West R if you made it this far thank you so much for watching and I’ll see you next time

FOR EDUCATIONAL AND KNOWLEDGE SHARING PURPOSES ONLY. NOT-FOR-PROFIT. SEE COPYRIGHT DISCLAIMER.