An AI Botnet

Roots push through dirt as mycelium spreads below. Hackers push through networks, finding paths nobody knows. Both move in silence, bypassing every lock until AI is weaponized and they join my cyber flock. What? Your box is mine now. Okay, this is a online forum post, hack forums, bulletin board, whatever that is. It's weird. It's not the dark web. Obviously, it's on the clear. It's not aion domain, but a lot of weird people hang out there, you know, doing weird cyber crime stuff. This is a post for Mycelium, an advanced C++ multi-exploit framework. They say Mycelium is a sophisticated cross-platform C++ framework designed for advanced security research red team operations in distributed computing. It's designed to be a featurerich realorld botnet application. They say all this in this paragraph like presuming, oh, it's for research, it's for education. But then they go on, unlike typical research tools, mycelium is a production-grade threat actor capability with full implementations. They straight up say, "Hey, this is for cyber criminals." Okay. All coming from sis call hex 3C. Apparently a noob on hack forums with one star. If it's not already obvious, I think it's pretty clear this is an AI generated posting. You got some like unformatted markdown indicators like the header hashtag and the inline code blocks. But there's some kind of interesting stuff here. They say all these big exploit kits with, you know, CVE from 2021, but exchange proxy shell and log for shell might be in there and some like usual enterprise hits. They say this thing is crossplatform. They have persistence with Windows. A little registry runes one schedule test. Hey, okay, we're just like scratching the surface here. Stealth and evasion though. Okay, we are doing some runtime patching of AMSI, the anti-malware scan interface for a lot of Windows and Defender oriented stuff. Those are all real things, but I'm just looking at the post, right? This online advertisement. And as much as I'm poking fun and making light of this thing, there's kind of an interesting idea or concept in here though. Uh, distributed AI grid like the mind collective. First paragraph is pretty dumb. A capability envisioned and developed by the author. Okay, thanks Chat GPT. that transforms the botnet into a distributed AI inference network for advanced social engineering and autonomous propagation. So allegedly a shared resource pool where infected nodes or bots that have AI access with Olama or API keys could be using like hijack copilot sessions act as service nodes. All part of this botnet, right? Context aware routing. They intelligently route tasks based on priority. High-value targets to find specifically like GPT. Why? Why? Why GPT4? We're on like 5.6, guys. Clawed API keys. A little fable. A little mythos. This one's neat. routes mass spam fishing to free local models. Uh, so there's no cost there. And then a hyperargeted fishing engine to analyze victim email and chat history to mimic their writing style, tone of voice, and context for undetectable fishing campaigns. Little bit of business email compromise. Might be kind of wild if it came like from your real persona and identity with your style of writing and communication. and messaging propagation hijacks Discord, Slack, and Telegram sessions to send AI generated trust exploiting messages to contacts. Some of these might be wild ideas, though. Injects prompt traps into code docs that cause victims AI assistants like GitHub Copilot to hallucinate malicious install commands. That's just malicious skills. You can say it. That's just prompt injection. Now, I was notified about this by my friends over at Flare. They had a blog post and a writeup uh talking mycelium framework first ever witnessed AI as a service botnet. I think this is still a really interesting idea and concept. More often than not, I think a lot of the news and headlines that you tend to see of like, oh, AI fully autonomous exploitation frameworks, AI powered malware evading EDR with AI, a lot of that I feel like is is hype. It's not reality. It's very clear and very obvious that AI does streamline speedrun and just give you a bit of superpowers both offense and defense. Like anyone can harness this, both good and bad. And hackers, threat actors, and cyber criminals are very obviously taking advantage of that capability. But we haven't gotten to Terminator, Skynet, Zagod, God mode, AI, robot, overlord, killer, cyber crime. You know what I mean? And what we're seeing here with this mycelium thing, again, we don't have any source code, any proof of concept. There is no actual implementation. There is no proof. And I'm just look going off of a post on a forum. So I can't say with any certainty if this thing is real or not. But at the very least, the concept, the idea is pretty real cuz imagine info stealer malware, ransomware, crypto miner, whatever sort of payload or implant, any access and infiltration on your computer. If it's going to go look for your claude or GPT, your codeex or open code or cursor or whatever Gemini robot gro crap, it could totally take advantage of that and use your system, turning your computer into the computational resource and AI infrastructure that for the adversary is going to be completely free and still be costing you money. And one thing that sticks out to me that could be wild, right, with this idea and concept, AI enabled, agentic, whatever, is that that could very well just be like one lane or part and piece of the rest of this capability. Like usual, okay, actual post exploitation or genuine exploitation, initial access, lateral movement, persistence, privilege escalation, blah blah blah. the entire rest of the cyber kill chain could then just be augmented and supplemented with AI. But hey, real quick, I do want to give some love to Flare because they are the sponsor of this video and both them and I have been having a lot of fun kind of exploring what are these like uh advertisements or the the public postings of what people are selling, buying and selling in the cyber crime marketplace and ecosystem and industry that this is. Now, if you are not familiar, Flare is the identity first threat intelligence platform that collapses the gap between detection and remediation. So, you could actually stop breaches and incidents in real time. They're tracking real threat actor communication across the dark web, like ransomware leak sites, like Telegram chat groups, like information stealing malware logs to find cookies and passwords and credentials, and literally the front door that genuine cyber criminals use to breach and break into your organization. In that moment, whether it's 2 a.m. on a Saturday and you're not in the office or folks aren't online, they can automatically lock down and quarantine and protect that account, that individual, your teammate, your colleagues so that they aren't an entry point for a broader, bigger incident. I want to sign in and show you the platform for just a moment. While you can create your own different identifiers for your teammates, for folks maybe involved in your entra ID tenant and environment, you could be tracking supply chain risks from downstream ransomware events. You could be looking up exposed credentials, cookies, and all of the wild things that they are siphoning up all the time. You can see absurd like five and a half billion events, probably even more when we bring this to all time. But they have these wild sources that they're pulling a lot of insight from. You can see some of the places where they pull from here, even like named breaches that are in the news and in the headlines. But I really love getting to the events tab because the global search almost kind of gives you like a Google for the dark web to be able to look up, query, and search for anything across all of those different locations. Let me unselect all of them and then just get to these forum posts and I'll apply that. And I want to look for that myelium. Yeah, here is that exact listing that they had a screenshot of inside of the blog post. You can see a lot of the metadata. They do include the raw URL even for a lot of the dark web like onion URLs. That's wildly cool for research. And you can get the full contents. Maybe they extract out the images. You can get the conversations again in Telegram in these info stealer logs, ransomware leak site data. It's just a treasure trove of thread intelligence. Look, you know, I'm a fanboy. I think Flare is just ridiculously cool. They always have awesome research to just really neat insight on what real threat actors and cyber criminals or adversaries are doing, what they're chatting about, the tools that they're using, and that threat intelligence can better arm you, your organization, your company, your business, your team to prevent breaches and have the information advantage up against cyber crime. I hope you take a look at all the things that Flare is up to. There's a link below in the video description for you to be able to jump into a free trial. And big thanks, special shout out. Always appreciate Flare and their support helping this channel do all the things that it does. Thank you so much. So, back to our hack forum sis call hex 3C mycelium framework thing. Hey, here's an interesting other idea. AI and autonomous operations with a little bit of a DLL that they might inject. You have the capability to just ask your robot local AI model or some of the hijacked API keys, send a regular prompt, or you could ceue up some social engineering attacks, control the autonomous vulnerability research loop with exploit DB, GitHub, maybe look for specific CVEes, generating a payload with AI powering that. Again, I don't know. I can't say if this is real [clears throat] for sale, serious inquiries only. Okay, some of this is just cringe. But the thing is this could be real. Like even the text of this thing alone, you could probably copy paste crap into Chad GPT and it could make something like this. But the idea is real. And I think the idea is totally plausible, right? You know, in the AI era now, the execution is not so much the hard part. It's okay, having the idea and orchestrating it and architecting it in a way that is actually valuable. Even as dumb as it is, like the stuff that's in here could be pretty well turned into a prompt to make something like this. And the concept of a modular botnet, well, wouldn't it be wild if AI were sprinkled in just a smidge? It's obvious to me that we're going in that direction. And this post on its own is still just another breadcrumb and puzzle piece to say this is where the world is going. The Flare blog post and writeup has a couple cool little articles a little bit more tactical and might offer some good education and more detail on some of those specific concepts that could be bundled into this sort of mycelium framework. I'll leave all the links for those in the video description again for you in case that's valuable. What else is uh sys hex 3C up to? Oh. Oh, let's try that again. Oh uh no, I don't I don't I don't want to I don't want to do any of that. Nope. Nope. Never mind. This site is too silly. Wait, I want to be 5x per lead. Anyway, I just really like Flair's wrap up here and that look this notion of an AI as a service. You know, we've heard SAS software as a service, RAS, ransomware as a service. What is a ass? But there is a reality where infos stealer malware is now going to be ripping up your claw or codeex or GPT access. using your tokens and leveraging that for continued social engineering, content poisoning, prompt injection, and utilizing this as a botnet across a mass amount of victims would be wild. And they straight up say, "Look, a whole lot of this, 1700word post looks like a wild exaggeration. It's an AI generated post with marketing fluff and unreasonable capabilities. But the parts and pieces that it's referencing, we already know about like that. Those are things that the industry has seen forever. AMSI, those persistence mechanisms. So, what is new for the headline sake? For what? Everybody's running around like a chicken with their head cut off. Whoa. Wow. AI is crazy, huh? The craziness. We've seen all of that before. poisoning documentation and source code, stealing resources, but bundling that all up into a potential toolkit, a framework. Uh, okay. Again, big thanks to Flare and all their support and helping me with this video, giving me something to scream and shout about. Link below in the video description. Please do give them some love. But as AI adoption accelerates and computational resources become increasingly valuable, it's reasonable to expect like we're going to see thread actors explore similar models that will monetize compromise infrastructure as distributed compute platforms rather than just botn nets. Wait, wait, wait. Where is it? Where is it? Where is it? Where is it? Where is it? Your box is mine now.