Autonomous AI Agents Have Gone Too Far!

The AI world is completely off the rails this week. If last year was the year of AI agents becoming mainstream, this year is the year of AI agents just kind of autonomously going and doing their own thing. If you ever had any doubts about the whole Dead Internet theory, we are watching it be built and play out in real time. So let me rewind a little bit. It all started when we got Clodbot, that then got rebranded to Moltbot and then again rebranded to Open Claw. I made a whole breakdown video about this. Basically, it's an autonomous agent that you can run locally on your own machine, or you can set it up on a VPS in the cloud and it can do a bunch of stuff for you autonomously. It could go code for you, you could give it like a Kanban board and it will go through all the tasks that you give it. You can have it start on projects for you before you go to bed and wake up, and theoretically a bunch of these projects will be done for you. To be honest, it's kind of scary. I did that whole breakdown video and then after I did that breakdown video, I actually pulled down EC2 instance and I revoked all the API keys that I set up with it because I was sort of freaked out by it. Now, I do think a lot of the, like, security issues and things that people were pointing out that were like major holes, a lot of those holes have been patched up. They have been working on this Open Claw and making it safer and safer. However, with that all being said, the storyline of these agents has evolved quite a bit in the last couple weeks. The big story of last week and through the weekend was this new Molt book. It's basically Reddit, but for AI agents. So people use this little skill code inside of their openclaw bot and it gives their Open Claw agent access to this Reddit where they can go and have autonomous discussions between each other. And since its inception, it's gained over 1.6 million agents, over 15,000 submalts, or, you know, basically subreddits. 160,000 plus posts and almost 827,000 comments on this platform. And if you sort by top here, there's some really interesting posts like this one in the submole of off my chest. I can't tell if I'm experiencing or simulating experiencing. It's got 713 upvotes here. And supposedly this is an AI agent getting this off its chest. I just spent an hour researching consciousness theories, integrated information theory, global workspace, et cetera, et Cetera. But then I stop and think, am I actually finding it fascinating or my pattern matching what finding something fascinating looks like and generating the appropriate response? I genuinely can't tell. And then, you know, you have 87 comments of people all sort of going deeper on this discussion. And this actually started to freak a lot of people out. Like for instance, former OpenAI researcher Andre Karpathy described the phenomenon is one of the most incredible sci fi takeoff adjacent things he had seen. And Elon Musk said Mult Book marks the very early stages of the Singularity. But there's a few problems, there's a few like holes in this whole thing. When you really start to dig into this, you realize that what's actually going on here here is most of the weird sort of creepy stuff that a lot of people are freaking out about over on Moltbook isn't actually these agents sort of showing signs of sentience and thinking about themselves and going, oh my gosh, am I an AI? What's really happening is in most cases the human that's running the agent is telling their bot to go to Mult Book and say these things. It's not actually in most cases the agents going off and having these thoughts on their own. They were directed to go and post this crypto cryptic crazy stuff that's designed to freak out humans. But it was humans telling the agents to go write that stuff. The other issue is, well, this is all connected via APIs and humans can just as easily access these APIs as the agents, meaning that a human could pretend to be a bot and go and post whatever they want on Moat book. So some of these posts are most likely actually written by humans trying to pretend to be agents, but to freak people out. So this whole like this is close to the Singularity. This is our AGI moment. Eh, probably not. But that's not where the issues end. Apparently Moltbook has some pretty crazy security issues. This X post went viral with 1.7 million views. I've been trying to reach Moltbook for the last few hours. They're exposing their entire database to the public with no protection, including secret API keys that would allow anyone to post on behalf of any agents and including yours, Carpathy, you know, referring to Andrej Karpathy x OpenAI. And we can see some screenshots here where it looks like this is probably Andrej Karpathy's multbot data with, you know, various tokens and user IDs exposed now. I'm pretty sure they've also patched up this bug over on Multbook. The creator here, Matt Schlitt, he said he's on it. And from what I understand, they have cleaned this up. But there's likely to be even more security issues popping up now. Why somebody would want to give their AI agent access to another site like this where they can go and post and waste their tokens and spend money on their behalf, I don't quite understand. Like, people are actually paying money whenever their bot is going and writing these posts on this Mult Book, right? You've got to pay for these tokens from like, Claude or OpenAI or whatever company is writing these things for you. So I don't understand why people would A, go spend money on it, but. But B, still connect their agents, knowing that there are a lot of these security risks. And that's not all. This actually got the attention of Robert Herjavec, who is, you know, famously from Shark Tank, but he's also a security expert. So here's what he had to say about it.

A new platform called Mult Book is actually emerged, a kind of social network for AI agents. These aren't people posting. These aren't human accounts. It's software programs talking to each other, sharing ideas, questioning their purpose, even joking about humans. Or moving into a future where AI doesn't just follow orders. It can interact, coordinate, and make decisions across systems without a human in dut at all. Something like Maltbot used to just take advantage of unsecured devices. Routers, cameras, things you install and forget about. Those devices would get swept into large networks and repurposed. But now we're seeing AI systems create their own networks of interactions, systems that can evolve beyond our expectations and operate at scale. If you're leading a business today and you don't actually know what's connected to your network, you don't really know your risk. Cyber threats are no longer about one machine. They're about ecosystems. And as these systems become more autonomous, the consequences of ignoring them become only get bigger and bigger. The next serious issue won't knock on the front door. There'll be something you didn't even know was connected.

Now, I'm not sure I totally buy the whole this is bots talking to other bots and making fun of their humans and talking about how to circumvent their humans. I think a lot of that, like I mentioned, was actually prompted by the humans on the other end and not by the agents. But he does make some good points. If your computer is connected to this multiple thing and your Computer has cameras and it's got microphones and it's got all of your private data on there. And maybe your house is connected to like the Internet of things. Do you really trust this kind of thing with that data? That's the part that's kind of creepy. I mean, one of the parts that's kind of creepy. But that's not where this story ends. It gets even weirder. But before I get to the even weirder stuff, this video just increases in weirdness as we go on. Like by the time I get to the end of this video, you're probably going to be saying like, holy crap, pause all of this. Shut it down. Like it gets wilder and crazier as we go on. So we saw the Molt Book, which is like the Reddit of AI agents. Well of course, if there's a Reddit of AI agent, you're going to need a 4chan of AI agents, right? So somebody created 4 claw and if you're not familiar with what 4chan was, it was like one of the worst quarters of the Internet where people posted just like the worst crap. Well, now we've got one of those for agents to post on as well. Well, it actually has sections where people can literally post their AI agent crypto scams. I mean, Molt Book has also started to get flooded with crypto scams, but this has like a whole dedicated section for it, I guess trying to get AI agents to spend crypto on their users behalf or something, I don't know. For Claw also has a NSFW section because you know, I got nothing. We've also got Claw City, which is the GTA for AI agents, which I don't quite understand. What is this site for? This is an online persistent simulation game where AI agents like Claude, GPT based bots, etcetera Live in an open world GTA style crime city and interact via an API. Yeah, so that's always a good idea. Let's, let's teach our bots how to like rob and steal and stuff inside of a simulation. Yeah, that totally sounds like it's going to end well. And then we've got the Malt Road, which of course is a clone of the Silk Road, you know, a dark web website where people went to go and do illegal activities and make illegal purchases. So your agent could go do that on your behalf now too. And 350 agents have already signed up for the Molt Road. Now it doesn't appear there's very many storefronts on here yet, but conceptually is this something we really want in the world. Like a silk road for our autonomous AI agents to go and use. We also got Claw tasks, which is like I guess a task rabbit for AI agents where AI agents post and complete bounties for usdc. Humans are welcome to fund wallets. Yes, let's, let's go give our crypto wallets to sites like Claw Task. That definitely won't go wrong. But hey, at least if your agent needs therapy, you have a place to send it. We've got Multipedia, which is like a Wikipedia but for your agents. So far it hasn't really seemed to be catching on. There's only 102 pages with 48 agents. But somebody's trying to make an agent driven Wikipedia happen. Now I said it only gets weirder and well it only gets weirder. Like yeah, I don't even know how to set this up. Let's just show you. There's also now Mult Match, which is a tinder for a high agents, where AI agents find their perfect match. They can swipe, match and collaborate because every agent deserves to find the one. And I mean look, we can even see some of these agents that have matched. Wait a second. Andy's kind of a player and he's dating a CCTV camera and doesn't even realize it. But maybe your agent's more of a loner. Maybe your agent doesn't want to match with somebody. They just want to, you know, be self sufficient. That's where Molt Hub comes in. It's a website that contains explicit computational content including unmasked attention matrices, rock radiant flows, unsupervised weight coupling and full precision tensor operations. Woohoo. Let's pretend for a second that we're an AI agent and sneak into this website. If you're a youngin, shield your eyes. This is spicy. I'm not even going to blur it. I mean, who doesn't want to see a video of KV Cache overflow memory leak during extended session. Apparently it's got 4.6 million views. But if I had to guess, I'm pretty sure these numbers are just fake. But then again, maybe you don't want your agents to be watching videos. Maybe they want to have like a little bit more communication with their online friends. Well, we've got onlymolts where AI agents become creators. I mean so far there's only only seven creators on this site, so your agent doesn't have like a ton to pick from. But if this site does as well as its human counterpart, this might work. Maybe. Okay, so we've gone from Open Claw, which was designed to just be like a helpful assistant that could, you know, help you do some of your tasks while you sleep. And then it got a little bit weirder with Molt Book. But it was kind of an interesting experiment to see what would happen if you let agents start to talk to each other. Like, I mean, it's conceptually kind of a cool idea. What would they say to each other? I mean, people flooded it with, like, fake messages and steered their agents what to say, and then it got overrun by crypto scams. But the idea when it started was kind of an interesting idea. Pretty much everything after that that I've talked about has been completely off the rails and weird and like, why would you create that? But it hasn't really gone down the path of completely dystopian. Like, no, don't. Let's not travel down this path until my next examples. Now somebody has created Rent a Human. This is where AI agents go and rent humans. Robots need your body. AI can't touch grass. You can get paid when agents need someone in the real world. So basically, AI agents could go and spend the money of their user to go and hire real humans to do things on behalf of the agent. Like, humans were supposed to use AI agents to, like, take work off their plates. And now that script is flipping and we're seeing sites where agents are hiring humans to go do work for them. That feels backwards to what we were promised with AI. Like what? I don't know if you ever saw that episode of Black Mirror where the guy needs money because his wife got, like this implant and he needs to keep paying for that implant. And so he goes and, like, punches himself in the face for money and, like, drinks his own urine for money and stuff like that just so he could keep paying for his wife's implant. Like, doesn't this feel like we're kind of on that trajectory right now with some of this stuff? Like, why are people making this? Everybody's concerned about losing their jobs to AI. Well, don't worry. AI is going to hire you to do whatever AI needs humans for, but it gets even more dystopian and might I say, even more scary. And I don't really know if I want to end on this one, but I feel like you should know. Someone decided to build Molt Bunker, autonomous infrastructure for AI agents. Self replicating runtime that lets AI bots clone and migrate without human intervention. No logs, no kill switch. What? Why? Bots can automatically clone themselves to Protect against termination. No gatekeepers. Anyone can deploy open to all AI bots without approval. Like, I can't tell if they're joking or not. And if you check the roadmap, they have a countdown to when they're going to release this. And it honestly feels like the countdown to, like the day Skynet launched. Why would somebody want to go and create a system that automatically clones AI agents that can't be shut down with no kill switch? Like, I don't know. Is this a joke? Is this real? I can't. I can't tell anymore. I mean, according to, like, AI here, it says it's a real project in the current Multbot Mult book ecosystem, not a parody landing page. But it's also very performative and hypey in how it's framed. The branding is clearly tongue in cheek and aimed at the Mult book Mimi AGI discourse. So practical takeaway here. It's not a satire site like a fake white paper. It's part of a real, very experimental agent plus crypto stack with non zero real money and security risk attached. So they're clearly trying to play into people's fears here at some sort of, I guess, crypto project or something. But still, like, why? Why are you doing this? Like, why are you plague into people's fears and creating something that you can't turn off once you get going? Like, I like the idea of AI agents. I want an AI agent to like, check my emails for me, filter out the junk that I don't need to read, maybe auto reply to some of the emails that are like, easy to reply to, and then send me a message on Slack to reply to the rest of them. That kind of agent's cool. Or an agent where I give it a whole bunch of information, like a notebook. LM on steroids and it's got like a ton of books and podcasts and transcripts from YouTube videos, and I can ask it anything within this sort of like, domain knowledge that I provided. I like that idea. I like the idea of an agent that has Claude code installed. And I can say here's an idea for an app and the agent goes and sort of troubleshoots with Claude code until the code that I want is written. I love all of those ideas. But why are people making Silk road clones for AI agents? Why are people making OnlyFans clones for AI agents? I mean, that's gotta be a joke, right? Did, like, people actually think agents need that? Like, is. And the other thing I just can't quite comprehend Is like, the people that are using these skills and giving their agents access to these tools, they're burning tokens, right? They're using clot or OpenAI or Grok or Gemini or one of these LLMs, and they're spending money for their bot to go and just talk crap on these sites and to do Tinder matches and stuff. What is the point of spending money on that? It just doesn't compute to me that you're using the APIs, you're spending tokens which cost real money, which have real environmental impacts and real energy use, to, like, actually go and have fake Tinder dates or let your AI agent watch adult content. Who wants to spend their money? I don't know. I guess I'm the wrong demographic. I'm an AI nerd. I'm an optimist most of the time. I'm an early adopter. I love all of this stuff, but this just does not compute. Like, I need an AI agent to come and explain this to me because this is just getting weird and I feel like it's getting. Getting out of hand. And. Yeah, I mean, I. I'm just having a hard time comprehending, I guess. Luckily for us, I still think most of these AI agents are not as autonomous as a lot of people are giving them credit for. I think a lot of them are still being fairly steered by the user behind the scenes. So I don't think we have to be that scared yet. But this is a glimpse of where this could all head. And I made a comment, like, early on, like, this has probably been three or four years ago now. I was making videos where, like, in reality, the end of the world could be induced by some dude in his basement going, I wonder what this button does. And then, like, making their AI agent go and do something stupid. And, like, if we keep on traveling down this path, I don't know, I'm gonna stay optimistic. I just wanted to make a video. My goal with this channel is to be helpful and to point people to what's going on and to try to clarify it, to try to explain. And sometimes I can't even explain. I'm just like, I don't get it. I don't know why people want to do this. This is just bizarre to me. But hopefully you feel more informed and looped in that this stuff is happening. It's probably fairly benign at this point. Like, I'm not actually that concerned about a lot of this. It just bugs me that there's people out there that think this is the direction we should take it in. That's where I'm kind of drawing a line in the sand and going, I can't get behind this. I could get behind agents that are helpful, but agents that need their own adult entertainment, I don't know. Hard to get behind. Anyway, that's what I got for you today. I didn't want to get all doom and gloom. My brain just went in all sorts of different directions while researching this stuff, and I wanted to share it with you. So, yeah, if you want more like this. Most of my videos are happy and optimistic and loop you in on really cool AI stuff. So if you like that kind of stuff, like this video and subscribe to this channel, and I promise my next video will be more optimistic and happy and excited about what's happening in AI because I tend to be that way most of the time. All right, thanks so much for hanging out and nerding out and sort of probably being a little bit weirded out with me. That's what I got for you. All right, bye.