[00:00] Major leading AI lab in Thrappic has just warned that there are at least 12 different warfare applications for their AI being used right now.
[00:15] And that includes autonomous killer drones. In this, we're going to break down each of those 12 and then I'm going to give you my opinion about how existential this is
[00:28] because obviously the AI is coming to kill us all. Let's look at this. Number one, guided missiles. Clawed, it has been discovered. And these are all things Clawed has actually been used for
[00:41] and that they're investigating and have blocked. So the point is, this is what people are using AI for. And number one, Yemen-based. So think Houthi's taking over the Red Sea, funded by Iran, potentially.
[00:57] Yemen-based guided weapons engineering cell. This guided missile work would include systems architecture, implementation, build-out, testing, flight verification, and validation.
[01:12] In fact, in the notes here, they make the argument that this would create a guided rocket that uses a phone to help guide the device,
[01:24] And you essentially attach it to a ballistic missile with a range of potentially 2,000 kilometers. Kind of wild. And they used the technology to help determine control and positioning for the software,
[01:38] control settings, the firmware to build the pipelines, and performing flight simulations. That's just example number one. The people did end up test firing a rocket, but it looks like it ended up failing.
[01:51] This is also kind of interesting. It's like going back to the AI, like, hey, it didn't work. It blew up. Number two, torpedoes. Apparently, researchers in China are trying to create torpedo interceptors.
[02:05] So if somebody fires a torpedo, you get intercepted with a different kind of torpedo, like a Patriot interceptor, but for torpedoes. As well as anti-submarine programs. This in China via potentially a document written to win approval from a Chinese defense manufacturer.
[02:21] This is wild. This is like a 154-page report here, so there's a lot in here. We're just scratching the surface. This one was probably the scariest. Killer drones disrupting a Russia-based operation,
[02:35] using Claude to engineer an autonomous military drone swarm. What? We identified likely freelance Russia-based threat actors
[02:47] who set out to build a full-stack, autonomous, first-person point of view kamikaze drone system So in other words AI killer drones were being developed using AI that exists today Take a look at this An onboard language model would govern the attack
[03:12] steer the drones, and issue the call to detonate. The actors designed the platform for autonomous lethal engagement, including person-target class.
[03:25] So, in other words, what's the race of the person, the size of the person, the sex or the age of the person, visually classifying people as either a target, or maybe what they're wearing, right? So you could potentially say clothing as well.
[03:38] But targeting people based on how they look, and then autonomously deciding, you know what? That person is going to die. And having an AI govern that attack.
[03:51] That's crazy. And apparently they are using scraped Ukrainian combat footage to determine if someone is friend or foe. That's crazy. Then we've got weapons systems as part of this autonomous drone,
[04:07] loitering munitions, interceptions, strikes, UAV guidance, firmware, everything basically being built out with artificial intelligence. This one's a little more boring, but it's basically a way of saying,
[04:19] how can we get military weapons and then disrupt or cover up our tracks for acquiring the supplies and components we need for that? A, or B, how can we target our enemy's supply chain?
[04:34] Directed energy. China researching basically directed energy weapons and the supply chain related to those, including microwave generating devices,
[04:47] mapping out where suppliers are, countermeasures, and essentially laser weapons. How about bioweapons? Oh, yeah. making existing pathogens more dangerous?
[05:01] Yeah, we got that as well. What kind? Oh, how about this one right here? We, in this case study here, see that a biological safety classifier
[05:14] blocked a request for gain-of-function research into the Chikungunya virus, a mosquito-borne virus that causes debilitating systems
[05:27] such as severe pain and fever that can last for weeks or months. No licensed therapeutic purposes, so in other words, it's only designed to be evil. And because the Chikungunya virus circulates naturally,
[05:42] a deliberate release as part of a bioweapon would be difficult to distinguish from a natural outbreak. The engineer was trying to get them into infectious clones.
[05:55] and ultimately try to make the virus progressively more harmful as it repeatedly infected animals. Essentially, the data function argument.
[06:09] Next you think that the end of it Nah we gotta get some bird flu The LLM platform was used for researchers trying to get gain-of-function research
[06:21] in highly pathogenic avian flus or bird flus Related influenza variants are a prominent concern for their pandemic potential
[06:36] Great. That's just what we need to deal with in a pandemic. The near zero, what is this? There is a near zero population immunity in humans. And when spillovers occur, the effects are fatal in roughly half of human cases.
[06:53] Despite the high lethality, viruses do not spread effectively or efficiently from person to person. And so they're basically using AI to try to engineer these to be more virulent amongst, you know, transmission between animals and humans.
[07:11] Not great. But no, we're not done yet. We got the chikungunga virus, we got bird flu, and then we have orthopox, which is a variant of smallpox and monkeypox, apparently.
[07:24] the people researching orthopoxviruses decided to use randomly generated email addresses shortly before using Claude through probably VPNs
[07:37] and anonymizing infrastructure and they were able to run their entire operation in about an hour so it's like they could get in get the information they need and then get out Claude supplied the user with the information
[07:51] and thus was not blocked by our classifiers. Oopsies! Crazy. But we're not done yet. We need toxins.
[08:03] Yes, toxins. How can basically toxins, like those used in Botox, be used in dosages or variations that would essentially create
[08:16] incapacitating agents? Right here. Potential, like, they're good, maybe as a painkiller or therapeutic model, or treating migraines. However, there's also been this goal that someone was trying to accomplish
[08:32] of making this more paralytic, basically incapacitating people. So utilizing venoms and toxins to incapacitate people. Then, of course, you've got the classic human-directed exploitation of cybersecurity.
[08:49] And then, of course, you have espionage, using agentic swarms to collect as much information you can to harvest credentials, to surveil or to monitor or otherwise.
[09:04] These are insane and scary uses of artificial intelligence. And to me it all comes down to one thing I don think that the AI this is my opinion I don think it the AI or the killer drones that are the problem
[09:27] Killer drones are a problem. But I don't think the AI is going out there trying to say we need to kill all humans. The problem is this right here. It's humans trying to kill each other,
[09:40] using now a better rifle, so to speak. We've now gone from the flip-flops to the machine gun, right? And so it always comes down to this, doesn't it? But, yeah, this is scary, and it does, in my opinion, suggest,
[09:57] or maybe it helps explain why cybersecurity stocks have done so frickin' well lately, because this stuff is crazy, and it's intense, and it's probably only going to get worse,
[10:10] especially since you've got a lot of Chinese distillation that's occurring to take the information from these models and then apply it cheaply to their models without guardrails.
[10:26] Take a look at this. It's not so much, like, Claude can block them eventually, even though it fails many times in this example.
[10:38] But if Chinese models distill clod, let's say, then the guardrails may be removed, right?
[10:50] So you take all the insight over time by farming all of this and then you remove the guardrails. That's not great. And this is, of course, where they talk about distillation as well, which once again, that's a problem.
[11:04] So a lot of issues that are going to come around how does this get regulated in terms of, or like guardrailed basically and there are reasons to look at this and say wow this is scary tech
[11:17] but to me it's, the biggest threat is the humans and the war element. Like of course the United States is going to use AI and bombing the Iranians
[11:29] and the same is going to be true vice versa. Hey, we're going to use AI to target your ship, which is exactly what was just talked about, that quad may have been used to target U.S. ships
[11:41] in the straight-forward moves around the blockade to help enable Iran to strike U.S. ships. So everybody's going to use AI. So, therefore, it becomes even more important to have countermeasures
[11:53] and to try to prevent and stop war. It was probably that big. We get to where people have fewer reasons to kill each other. Because the ways to kill each other are getting crazier by the day.
[12:08] I don't even know how to advertise. These things each other feel. I feel like nobody else knows about this. We'll try a little advertising and see how it goes. Congratulations, man. You have done so much. People love you. People look up to you. Kevin, pass right there. Bye. Nice to meet you.
[12:20] And you too, but meet Kevin. Always great to get your take.