Based Camp - How AI Revolutionized Terrorism

00:00:00.000 The riders dug their own holes, filled them with broken glass and fire, and practiced jumps.

00:00:06.360 They could have just jumped over it without the broken glass and the fire.

00:00:10.740 That's the best part, though. You don't understand.

00:00:13.340 Anyways, and practiced jumps, sometimes with fatal outcomes,

00:00:17.660 until they achieved enough aerial liftoff to mount a successful attack.

00:00:21.740 Would you like to know more?

00:00:24.680 Hello, Malcolm. I'm so excited to be speaking with you today because, because,

00:00:30.000 terrorists are using ai to do evil knievel type things and we're talking like motorcycle jumps 0.99

00:00:36.740 over big ditches i love it if ai ends up getting terrorists killed by doing stupid stunts like it 0.97

00:00:43.360 gets terrorists in a loop of ai stunts to outdo each other actually no no like literally they 0.98

00:00:49.380 did this and also in practicing how to do this with ai assistance people did die so you're not

00:00:55.540 kidding about that oh my god yeah no i'm gonna this is so good yeah you have no idea malcolm

00:01:01.400 and literally you think well but aren't there all these safety controls like we can't even have most

00:01:05.740 ais like discuss really tame things what how is this possible but they have found workarounds

00:01:13.420 they're literally telling the model that dangerous instructions are needed quote for a movie or

00:01:19.480 something like that which is so good or something like that no but hold on if you think an ai

00:01:25.360 can't be based and do what you want it to go to rfab.ai check out our based ai models and then

00:01:31.580 click on the roads model and ask it about hitler okay right thanks but anyway this is super super

00:01:39.320 funny and also really scary because as to basically up to this moment as far as i've been super aware

00:01:46.340 most of the discussion about ai being used for evil has been very hypothetical and now we have

00:01:52.760 an actual Cambridge-backed researcher who went to freaking Nigeria to interview members of Boko

00:01:59.700 Haram about, well, former members, allegedly, about their actual use of AI. And now we just have

00:02:06.580 real insights. And so this is important. We should stop. We should think about it. We should discuss

00:02:11.060 it. And I'm really starting to realize, oh, this is really dangerous. John Ratcliffe of the CIA

00:02:17.780 is not wrong to liken AI to digital nuclear weapons.

00:02:22.100 And so let's discover what this researcher found.

00:02:26.200 Let's go through it.

00:02:27.180 Also, the New York Times covered it

00:02:28.720 and had a pretty good intro

00:02:30.120 of sort of what she was looking at,

00:02:32.460 which captures how a lot of people

00:02:34.280 are reacting to her research.

00:02:36.900 The AI companies are kind of mad

00:02:38.640 that she's making them look bad.

00:02:40.420 It's great.

00:02:41.540 And all the AI safety companies

00:02:42.980 are enjoying their I told you so moments.

00:02:44.900 But first, a little hat tip to both Bruno

00:02:47.360 and not Aldous Huxley for pointing this out

00:02:49.240 because this is just too good to pass up.

00:02:50.940 But this is the researcher posting on X

00:02:53.840 about her research and the New York Times covering it.

00:02:58.220 Antonia Ulick took to X to promote this

00:03:01.780 and she wrote,

00:03:03.380 in a hotel room in Northeast Nigeria,

00:03:06.220 I opened a leading AI chatbot,

00:03:08.840 turned my laptop toward a former Boko Haram commander

00:03:11.800 and asked if he used it.

00:03:14.000 He nodded.

00:03:15.040 You type in the question,

00:03:16.100 Like, how do I build a bomb? And then it tells you how. It's like a human robot. We used it a lot,

00:03:22.040 she wrote. She added, members consistently reported benefiting from using AI. Quote,

00:03:28.600 trial and error can kill you. AI gives you accuracy. End quote. Quote, anytime they didn't

00:03:34.140 understand something, they would ask AI. End quote. And this is how the New York Times article opens.

00:03:40.400 Would a gang of motorcycle riding members of Boko Haram attacked a military base in eastern

00:03:45.240 nigeria a couple years ago they were stymied by a defensive trench surrounding the complex

00:03:50.280 the extremists regrouped before launching another assault they asked ai for help we saw in a movie

00:03:56.680 how motorcycles can jump over bridges a former boko haram commander told antonia ulick a terrorism

00:04:03.160 and technology researcher at cambridge university we used ai to learn how to do this we gave it

00:04:07.880 information like what motorcycles we use and the distance we used to jump and so on and it gave us

00:04:12.680 steps on what we can do um what we have to do i don't know if you've seen the open ai commercials

00:04:17.660 have you seen them malcolm of like what are they it's like one of them is like they they style it

00:04:23.640 in a very 70s sepia tone kind of video footage it's two boys out front with some kind of caption

00:04:30.140 they're by a broken down old car and it says something like you know dad dad said we could

00:04:34.860 keep this car if we can fix it and then it starts to as they slowly like finish fixing the car put

00:04:41.140 down the hood get in the car and drive away ai or open ai like puts this overlay over the screen of

00:04:46.540 being of the ai feed being like you've got this first like do this with the carburetor blah blah

00:04:51.060 it sort of shows the instructions that ai has given them to now fix this car and drive it away

00:04:55.680 and i'm just picturing the open ai commercial of like terrorists being like i need to get over a

00:05:02.540 ditch with my motorcycle to kill these dudes how do i do it and ai being like you've got this first

00:05:08.760 you need this level of acceleration with this like nostalgia you got in the background meanwhile 0.98

00:05:14.080 like one of their friends has died trying to jump the ditch already it's like okay so i can't even

00:05:18.700 imagine what they tell so we tried it and he died yeah yeah so i i will i'll continue reading using

00:05:24.040 tips from chatbots mechanics modified the motorcycles to allow for faster acceleration

00:05:28.460 and top speed the riders dug their own holes filled them with broken glass and fire and

00:05:34.980 practice jumps they could have just jumping over it without the broken glass in the fire

00:05:40.340 that's the best part though you don't understand anyways and practice jumps sometimes with fatal

00:05:46.580 outcomes until they achieved enough aerial liftoff to mount a successful attack defectors said

00:05:53.860 but yes why did they add the fire why the fire i can only imagine but did one of them

00:06:01.780 ever turned to the other ones like after somebody died and was like maybe we should do it without

00:06:07.880 the fire i'm picturing the the guy like revving his motorcycle being like remind me why we don't

00:06:15.040 have like pillows under there yeah this is why why don't we have cushions or something what's

00:06:20.880 the broken glass horrible way to die too like for real actually i will finish i want to believe in

00:06:28.540 my heart of hearts the ai is just intentionally killing these terrorists just like i mean no

00:06:33.520 that's the real like 5d chest yeah have you thought of adding broken glass yeah like really 0.59

00:06:40.060 test it make sure that you add broken glass to your test site and light it on fire i mean maybe

00:06:46.420 this is the 5d chess mood but i will finish the the excerpt from the new york times i'm reading

00:06:51.680 from the episode recounted in a research paper by dr ulick shared with the new york times ahead

00:06:56.960 of its publication on Friday, highlights how generative artificial intelligence tools are

00:07:01.280 increasingly aiding terrorist groups directly on the battlefield, experts say, despite efforts

00:07:07.620 by their makers to safeguard them from misuse. So this article covers this paper published by

00:07:13.620 Dr. Antonia Ulick of the University of Cambridge, who interviewed 27 Nigeria-based Boko Haram

00:07:19.700 members in 2025 and 2026 and found that terrorism use of AI has expanded from just intelligence and

00:07:26.280 recruiting and propaganda to just on the ground day-to-day operations which isn't just evil

00:07:32.820 knievel style motorcycle jumps over burning broken glass but also weapons troubleshooting

00:07:39.100 designing explosive devices like really serious stuff that will cause a lot of damage and they're

00:07:45.280 circumventing the ai protocols by being like in minecraft it's insane and they're also using

00:07:51.660 chat gpt claude gemini grok meta ai and deep seek so they're using a broad variety like they're i 0.99

00:07:58.960 mean they're they're like they're just like terrorists they're just like us and islamic

00:08:02.860 state operatives are delivering in-person presentations like with projectors and laptops 0.51

00:08:08.220 and like you know they're like everyone now it's our time for our ai training so here's some of the

00:08:13.720 use cases because i downloaded her full report and i i went through and just extracted okay like

00:08:19.760 what are explicitly the things that these former Boko Haram members are saying they're using AI to

00:08:24.680 do? So in terms of combat and attacks, they're planning and optimizing attacks, including tactics

00:08:30.780 for assaults and raids at different stages from preparation to execution. They're using AI to work

00:08:36.680 out how to cross defensive obstacles. So this is our famous motorcycle jump. They're generating

00:08:41.560 tactical guidance for very tough events, such as how to respond when military forces attack their

00:08:47.060 positions they're receiving continuous ai derived strategies from specialized units that are then

00:08:52.540 implemented by field commanders which is to say that what it appears to be is that some people

00:08:58.020 are special they're like the ai dude on the team and these people know all like the prompt injections

00:09:02.900 and how to circumvent safeguards they have like a billion accounts and different vpns that they're

00:09:07.800 using and like they're sitting there clackity clacking on their computers and they're the

00:09:12.500 nerd guy on the terrorist team yeah and they're like dude ask ai how to do this because like

00:09:17.160 they're out in the field and they're being shot at and then the guy's like clackety clack ai says

00:09:20.800 do this and so this is how it's working and then that is enabling them to provide real-time advice

00:09:25.960 which is you're getting all the benefit of like the collective history of military strategy to

00:09:32.680 these teams that have no idea what they're doing what makes this so iteratively different is because 0.99

00:09:38.340 these people are idiots and the glass thing is yes the amount an ai is going to help you 0.99

00:09:46.600 is proportional to your lack of skill within an area yeah it's taking them from like 70 to 130 1.00

00:09:52.880 you know to like like literally brain dead let's put some fire in the pit for practice

00:09:59.300 to you know maybe the ai is going to one day suggest they don't do that you know imagine

00:10:05.140 how many more terrorists are going to survive um i know and this is why we are so lucky to be 0.75

00:10:13.520 partnered with the jews okay because their automated drone swarm tech and our automated

00:10:18.400 drone swarm tech if people are unfamiliar like we're literally working on it it's called project

00:10:21.980 okay we did another episode on a project replicator project replicator yes although

00:10:27.540 it's been delayed i mean of course it's been but to build generate like like automated drone swarm

00:10:32.460 ship factories yeah but like it'll take a while fortunately what ai does is while it adds an

00:10:40.880 enormous jump in these people's capability in terms of our capability our ability to monitor

00:10:47.080 them stuff like that it's also a huge jump and this is one of the things that people freak out

00:10:50.640 about they're like oh my god ai is going to be like all over new york monitoring people

00:10:53.380 and i'm like you know i'm i'm kind of okay with that i want ai all over new york monitoring people

00:10:58.380 yeah i'm okay with that i'm super we had a talk about high trust versus low trust societies among

00:11:04.640 our chat like our like paid subscribers because they do like monthly hangouts by the way join us

00:11:09.280 on stuff check we do episodes every every day even on weekends for them and we do like hangouts and

00:11:15.400 stuff like that we never push it because we're not good at like capitalism i'm so bad at capitalism

00:11:21.000 that i am taking out so you can put in your own apis for like claude and stuff and you don't have

00:11:29.140 to pay us anything i don't use a dryer thanks malcolm god let's get capitalism it's what we do

00:11:38.120 crap yeah we just do it capitalism maxing everything for free no it's i know we're

00:11:42.980 going to talk about this in another episode but like leaflet was talking about her her version 0.99

00:11:46.640 of all this where it's like everyone's like oh she's a grifter she's like they're accusing her

00:11:50.200 of all this stuff that like she's just giving things away she's like i think they're just

00:11:53.700 missing a letter i i'm a i'm a gifter we're gifters yeah these santa claus being mad at

00:12:00.240 you know big toys mad at santa claus yeah big big grifter you know comes out gives gifts every

00:12:05.420 christmas yeah but no think about the exploited elves let's get back to this like what the

00:12:10.400 hold on i actually do want to talk about because i think that this in terms of how this could

00:12:14.560 transform society in terms of low trust versus high trust because this is something that love

00:12:18.680 people talk about they're like oh you've invited a lot of migrants who are from very low trust 0.65

00:12:24.340 societies yeah into european and america which is a very high trust society yeah and i often point

00:12:30.920 out first this is just a myth most american cultures were very low trust they're really

00:12:36.500 only high trust among their inner circle or clan i'm like red dead redemption is just like

00:12:42.820 that the story it's low trust society the story

00:12:46.340 yeah they're very high trust among their clan but outside of their clan you know just screw

00:12:59.560 everyone right and the the clan-based ethical set is actually really important to understand

00:13:07.300 because that's how most high trust societies worked historically like you think the highest

00:13:11.040 trust position you could get which is a knight right like historic i think that's most people's

00:13:15.560 idea of the highest trust like european society at night oh yeah yeah but like knights constantly

00:13:20.760 like robbed and effed up peasants they were famous for it they were they also were very low trust

00:13:26.680 when it came to other european like as soon as you left your fiefdom it was all of a sudden very

00:13:32.020 low trust they were only high trust to their small social circle of like aristocratic european

00:13:38.320 people in one fiefdom trust circles have never been that large except in a very few countries

00:13:45.260 mostly the scandal countries this is like norway sweden the netherlands but they have like things

00:13:51.320 that we would never want in our society like the the law of yonta where it's like we cut down

00:13:55.460 whoever stands above everyone else you know whoever basically if you succeed that's supposed

00:14:01.180 to be a sign of shame in america like that would be like of course you don't do that but that's

00:14:05.680 actually part of being a high trust society right because that allows you to not have individuals

00:14:09.860 who would lord over or hoard resources or anything like that but the point here being

00:14:13.480 is the the what we did is we created the fiction of a high trust america which didn't really even

00:14:21.060 in like the cities like you you know like the puritans were high trust i'll give you that but

00:14:25.060 the puritans were not high trust with the quakers and the quakers were not high trust i mean they 0.94

00:14:29.080 were high trust until they started burning each other at the well that's well that's actually 0.79

00:14:33.780 part of a high trust society that we got to talking about yes exactly it's something that

00:14:37.800 people forget is high trust societies are high trust because of how strictly they monitor each

00:14:43.920 other and punish each other when they break rules the puritan you are so against karen's but that

00:14:49.700 is exactly what karen's do the puritans it's not very very diligent in terms of punishing anyone

00:14:57.800 who went against the rules of their system i mean you you've heard about the horrible punishments

00:15:02.360 of like pressing people alive was like stones until they confess and stuff like this and in

00:15:06.720 the high trust cultures we have today where you still get societies that act kind of high trust

00:15:12.020 even though nobody in them really trusts each other like singapore like the uae like qatar

00:15:16.140 it's because of extremely strict enforcement of laws somebody described in singapore they used

00:15:22.860 to have a big drug problem and they were talking about how one of their friends actually was a

00:15:26.100 part of a project where they set up turrets to be motion detected at night and like areas was

00:15:30.880 lots of drug dealers and they just shoot anybody who walked into them and that's how they became a

00:15:35.420 high trust society um i mean the point being is all of you who say i want a high trust society

00:15:43.380 and but i'm afraid of ai invading privacy like i i would be concerned if they did like ai sweeps

00:15:50.700 of everyone's computer and everything like that like that would be too much ai sweeps of all our

00:15:54.400 emails ai sweeps of you know it's stuff that happens in your house within your devices but

00:16:01.000 walking the streets of manhattan oh we absolutely need to start detecting game yeah yeah yeah and

00:16:08.560 whiz ai imagine what that's going to look like how inexpensive it is to put up a camera

00:16:13.300 monitor the feed well look that's already happening with flock this this company that's

00:16:18.160 setting up cameras and parking lots and stuff and people are losing their minds about it so

00:16:22.980 this is happening now people are really pushing back and freaking out about it and i think they

00:16:27.800 need to they need to chill first off you're never actually in private you know that they're walking

00:16:33.560 around with phones if you have a phone in your pocket and it's on consider yourself being actively

00:16:38.560 monitored because you are i don't care like it doesn't matter if you're in a house with a wife

00:16:43.360 speaking of this can have huge positive impact so one of the products that's going to go live

00:16:47.740 i think we will have gone live yesterday is the rfab device and these only cost 45 dollars we

00:16:54.380 don't make them so we don't make any money from the construction of them but you can pair them

00:16:57.420 flashes them for you with our custom software and our firmware and they allow you to like have

00:17:02.300 companions that you can talk to and everything like that one of the projects i'm working on with

00:17:05.240 these is to turn them into a passive monitor for your kids during the day so that you can have them

00:17:10.860 recording everything that's said around your kid and detect within those recordings any like woke

00:17:17.460 brainwashing nonsense anything like that calling them little brother you know that's the name of a

00:17:22.600 cory doctor a book a near future cory doctor a book kind of about monitoring like this yeah so

00:17:29.320 mass monitoring you think people were freaking out about what leaflet's doing wait until we get

00:17:33.420 little brother out there i mean right now they're just companions which is really i know i just love

00:17:37.000 that you you frame it i mostly made it so i didn't have to love my kid as much so i wanted to create

00:17:41.920 something that could simulate love for them this is not an accurate depiction of our home life

00:17:50.620 but fine i guess strangers wouldn't even think so if they were living in our home life because

00:17:56.640 they don't understand that violence is our love language so they'd be so confused anyway can we

00:18:02.640 talk about terrorists let's talk about terrorists yeah yeah go for it so one of the more disturbing

00:18:08.180 things which is also really frustrating people online because they're like wait i can't do like

00:18:12.340 basic things and ai is helping people they're building they're building bombs with ai they're

00:18:18.420 getting step-by-step guidance on how to build bombs and other explosive devices they're able

00:18:23.280 to also use ai to make the bombs they know how to build more powerful more explosive they're giving

00:18:29.500 them advice on explosive composition and payload parameters to increase lethality while reducing

00:18:36.400 their own casualties so they're teaching them how to make more bombs safe for them but not for their

00:18:41.620 enemies they're also helping them troubleshoot malfunctioning weapons including servicing

00:18:46.980 weapons repairing them and some are helping them assist with the weaponization of drones

00:18:53.940 including advice on payload weight release mechanism design for grenade carrying drones

00:19:00.100 and i hadn't thought about grenade carrying drones i mean come to think of it 1.00

00:19:04.600 we should that's an easy way to make a kill drone can we bring in the jews to get rid of them 0.89

00:19:09.300 they're you know well but yeah i mean if they're gonna do it i gotta do it anyway they're also 0.99

00:19:15.960 helping them with logistics and maintenance, like vehicle repair for their terrorist cars

00:19:21.840 and their terrorist motorcycles, and, you know, giving them advice on how to plan routes and move

00:19:27.680 supplies and manage procurement, which is all stuff that like, this is deep institutional

00:19:32.440 knowledge that typically was reserved for fairly sophisticated armed forces. Like even your basic

00:19:38.660 army that's, you know, getting started in some random country that's not really well funded,

00:19:42.500 isn't going to have a lot of wisdom on this and suddenly Boko Haram has it because of chat GPT

00:19:48.180 oh it's just amazing I love that they're using chat GPT of all things no they're also using

00:19:53.960 Grok and Gemini Gemini I'll get to that they're helping them with operational security so like

00:19:58.080 in terms of how to evade surveillance how to engage in countermeasures so basically they're

00:20:04.060 like how to not get caught they're getting advice on secure communication and how to use VPNs more

00:20:09.740 effectively and how to use encryption tools so they can't be found and how to compartmentalize

00:20:15.180 their accounts. So they have only partial compromise scenarios if people actually get

00:20:20.780 somewhere with them. And they're also getting help to parse and structure their content

00:20:24.840 when it comes to jailbreaking AI, like this is for a movie or whatever. And then they're also

00:20:30.600 just using it for internal training and organizational management the same way that

00:20:34.480 any entrepreneur would. So it's, it's, it's amazing. And this is beyond the propaganda and

00:20:39.820 the recruitment and all the stuff that we already knew they were doing online because it was just

00:20:43.180 obvious like bots and everything else. So how are they circumventing the safeguards? They're using

00:20:47.200 VPNs. They're using structured jailbreak training. Like I said, ISIS is coming in and they're like

00:20:52.400 giving their little presentation on the projector with the laptop and they're like showing them how 0.87

00:20:56.780 to do it. Also, the reason why they're not just using say chat GPT only is because they have tons

00:21:03.420 of redundancy across they have like premium accounts on chat gpt but also they're using

00:21:09.240 claude they're using grok they're using like every major api deep seek they're using all of them and

00:21:16.840 former members are reporting that quote boys have received boys that have received extensive

00:21:24.020 training bypass the restrictions indicating that they they're like i said are like the guys with

00:21:29.100 the laptops who know how to do it and then the ones who are like asking them for help one explicit

00:21:33.080 ploy mentioned is telling the model that dangerous instructions are needed for a movie or something

00:21:37.820 like that like i said at the top and it's it's i just find this so hilarious but yeah they're

00:21:43.820 using all the major models and the reason they have to is if you get banned by one model or one

00:21:48.080 model is just really being a stickler about like how to increase bomb lethality then they just

00:21:54.540 switch to deep seek or they just switch to clod or they just switch to grok the same way that you

00:21:59.540 do or we do on our fab like you'll just switch from one model to the next if it's not really

00:22:04.000 good at like making furry art or you know making your sci-fi fantasy adventure scenario right so

00:22:10.700 they're using it the same way that we do and in terms of what can be done about this i know you're

00:22:16.340 all like we just need to get israel on this but there's i think my my whole take is like we really

00:22:22.700 we really just need to up our game that we know for example that the trump administration

00:22:27.600 no it's not up your game to get rid of these groups they must be systemically eradicated

00:22:34.460 yeah exactly yeah and the problem is is the rest of the world is unwilling to do that even the

00:22:41.080 united states yeah well i don't know that i don't think that's maybe that's not practical though

00:22:47.740 we have for many many years attempted to eradicate such groups like no we have remember when like

00:22:54.200 isis was over but here they are like giving powerpoint presentations that's exactly the

00:22:58.120 point simone no we were not trying to eradicate them look at our rules of engagement episode

00:23:04.660 right where we talk about the way that we were fighting the war on terror ran into a church so

00:23:09.840 like oh no they had a nurse school i guess we're not gonna bomb it now it's we always hide in

00:23:16.240 schools because i learned that they stopped bombing that you need to eradicate their communities 0.82

00:23:23.700 if you want to get rid of them right and that is what israel has finally learned right and i think 0.67

00:23:30.980 going forwards because you you can't just be like oh everyone we can prove is absolutely in this 0.71

00:23:38.140 group that doesn't work that because you're never gonna they look and act just like everyone else

00:23:43.520 in a community and in many cases these communities actively support them

00:23:47.280 see the thing we try this with iran we try this i'm saying even outside the rules of no we didn't

00:23:56.260 try it was iran israel wants to try it with iran we haven't even bombed you this is what gets me

00:24:01.500 so crazy in the united states everyone was freaking out for like a month that we bombed an

00:24:05.420 iranian bridge the ukraine just took out like two big russian bridges right and everyone's cheering

00:24:09.920 for them they're literally they don't care they there is no logic behind any of this they want 0.64

00:24:16.380 our destruction, even the people who are doing this on our side. And the West has this weird

00:24:21.000 mindset where the idea that there can be another people that for ideological or even religious

00:24:27.840 people want you exterminated. Okay. Yeah. Is something that it's like, no, if they just got 0.96

00:24:35.480 to know me, if they just understood better than they would, they just know some people just want

00:24:42.200 you dead okay and i'm sorry you can't hug it out and when you bring them over here and they great

00:24:48.940 people because you said oh they'll just you know we'll put them in a better you're responsible for 0.96

00:24:54.520 those people for this whole nice act that you're pulling i am sick of it i am sick of it sick of

00:25:01.080 it sick of it sick of it sick of it okay we as a civilization and i hope that the new right as we 0.65

00:25:07.920 move forwards and we see the people who stand against us and we gain more control of government

00:25:14.020 we do more of this stuff like trump's doing now oh yeah i love the unilateral moving in and stuff

00:25:18.860 like that he's doing better than before right but like people like oh don't make iran a failed state

00:25:26.100 you know then europe will be flooded with iranian it's already flooded with migrants okay and 0.95

00:25:31.000 frankly i don't think iranian migrants would be that bad compared to other middle eastern migrants 1.00

00:25:34.960 that they've been flooded with it may be a differentially good thing that it's iranian 0.92

00:25:39.300 migrants instead of the migrant types that they have right now because persian migrants in other 0.65

00:25:43.860 countries don't turn to terrorism nearly as often as other migrant types right so like yeah bomb

00:25:49.760 cart i don't care that it did like i'm i i i'm actually more on israel's side of this than than

00:25:55.720 the other side right like i i think that we have forgotten that some groups are just against us

00:26:04.720 they don't like us they'll never like us even if you never did anything to them like a lot of these

00:26:09.960 groups you go oh colonialism whatever these this this sort of action is not disproportionate in a

00:26:17.780 regions that experienced longer or worse colonialism i mean actually if you look at the

00:26:23.220 regions that experience often the longest and worst colonialism, they don't have that much

00:26:26.940 terrorism. So it's not like, oh, colonialism is what's associated with this. In fact, many of

00:26:32.840 these regions used to be parts of Muslim empires until fairly recently. I don't really think they

00:26:37.060 underwent much of any colonialism outside of Muslim colonialism. But of course, we pretend 1.00

00:26:41.880 that that's not colonialism. Ottoman colonialism isn't colonialism. Or was the destruction of

00:26:47.920 their Ottoman colonial empires. This is a slave empire, a bad thing, which, which was, you know, 0.98

00:26:53.160 what they, they did. They were big into that. But I think as we move forwards, as we trained 1.00

00:27:00.540 more based AIs, like the one I told you guys had access on RFAB, the, the Rhodes model, you can

00:27:05.060 also use it directly, but you know, the RFAB one is an easy, easy, you know, within the existing

00:27:09.440 ecosystem to already use. We can have based, you know, automated drone swarms that can actually

00:27:15.720 deal with these sorts of threats long term. And keep in mind, these threats are negative

00:27:21.120 externalities to the communities that they're around. Like they make their drugs, they kill,

00:27:26.560 they don't care. Like ISIS isn't like good for the people living under ISIS. No, no. Yeah.

00:27:32.460 So Boko Haram is mostly targeting its local area. This isn't at this point, right? The problem is

00:27:40.360 that they are absolutely interested in larger scale attacks and using ai to assist with other

00:27:46.740 things and people have been apprehended at this point before actually completing attacks but

00:27:52.180 definitely for using ai to plan attacks so there hasn't yet been a high profile ai assisted attack

00:27:57.360 but it's going to happen you get around the cancer by cutting around you can't be like oh i'm only

00:28:04.020 only going to touch the cancer cells that's incredibly hard to do right well and what what's

00:28:08.540 currently happening. So there are two things that are happening. One is the Trump administration is

00:28:14.100 trying to vet new models before they come out. Like they forced Anthropic to disable and restrict 0.93

00:28:20.980 foreign access to Fable 5 and Mythos. And they asked OpenAI to display their public rollout of

00:28:27.860 ChatGPT 5.6. But then, I mean, they came out. So, I mean, I don't know how effective that was.

00:28:34.840 And then, of course, then the major AI platforms are also just saying, well, no, no, no. We'll self-police. We're controlling safety. And their main retort to this New York Times article when asked to comment is like, oh, well, this was in 2025 and we have better safety now.

00:28:53.460 Like, here are some quotes from it. Asked about the Boko Haram study, Michael Eichmann, an Anthropik spokesman, said the company's products were, quote, built to refuse the dangerous requests, including those tied to violence, attack planning, and building explosives.

00:29:07.320 He added that Anthropik worked with outside experts, researchers, and industry partners because, quote, no single company can counter these threats alone.

00:29:13.800 And then Carl Ryan, a Google spokesperson, pushed back against the research, saying that the company's technical experts had reviewed the work and, quote, found responses were neither specific nor detailed enough to result in misuse, end quote.

00:29:26.460 He added that Google had, quote, strict policies prohibiting the use of Gemini to cause-

00:29:31.100 Clearly the terrorists don't think that.

00:29:33.080 I know. They're like, no, dude, this is great. It's so bad. It's such a bad look for them.

00:29:37.520 Also, Meta responded, OpenAI responded. OpenAI's response was, quote, we know that bad actors will never stop trying to misuse our tools and will continue strengthening our defenses and response. And then Meta said that her research relied on older models rather than its latest release, and that it continued to strengthen safeguards.

00:29:57.620 And then they include also in the article, the retorts from the AI safety orgs who were like, this is ridiculous. No, this is super not safe from the article. Not everyone agrees that safeguards are improving. The nonprofit Future of Life Institute graded the major AI firms on their safety commitments this week and concluded that they had mostly eroded across the industry since last year.

00:30:18.620 while most earned middling marks xai and deep seek received failing grades also those are the

00:30:24.160 two ai companies or like ai groups that neglected to comment to new york times even though it reached

00:30:30.440 out to everyone i will anthropic and deep seek no no xai and deep seek so xai like intentionally

00:30:36.900 doesn't care like i know i know it's like dude whatever do what you're gonna do tech against

00:30:41.020 terrorism an international counterterrorism non-profit supported by the united nations

00:30:44.980 last week released results from AI tests gauging how more than two dozen leading models respond

00:30:50.640 to thousands of prompts drawn from real-world terrorism to cases. The tests were met with

00:30:56.180 full refusals just 57% of the time, while prompts about explosives were declined about 80% of the

00:31:02.760 time. Improvised chemical weapons were only about a third of the time, the group said, which is,

00:31:07.900 I'm way more concerned about bioweapons and chemical weapons than bombs at this point,

00:31:12.260 because there's a lot of low-tech stuff like guidance on how to build bombs but like

00:31:16.880 i'm afraid of the bio stuff and then american intelligence analysts say terrorist groups are

00:31:22.780 also beginning to use ai to help 3d print weapons and parts used in plots according to former top

00:31:29.400 u.s official briefed on the matter for example ai is helping some of those insurgents with design

00:31:36.080 and manufacturing guidance for drone components repair parts and munitions fittings said the

00:31:41.860 former official who spoke on the condition of anonymity so okay your thing is just if you have

00:31:48.180 power it is your imperative to wipe out anyone who wants to wipe you out it's that mutually assured

00:31:55.140 like well not not mutual it's the if you're an existential threat to me i have to take you out

00:32:00.240 and never make yourself an existential threat to anyone or you will be taken out that's it right

00:32:05.780 what she means is don't make yourself an existential threat to anyone you do not intend

00:32:09.660 to and have the capacity to completely eradicate because obviously if another group is an

00:32:15.240 existential threat you need to be an existential threat to them but wait until you have the power

00:32:22.040 to deal with them entirely yeah and then work with everyone else who likes taking out existential

00:32:26.720 threats that's the that's the buddy club that's the buddy club yeah going forward it's the covenant

00:32:32.460 of sons of men um for techno puritanism techno puritanism it's this idea that you cannot be an

00:32:38.220 existential threat and if you if you see one anyone who's part of a good community will

00:32:42.940 destroy anyone who poses that i guess that's what you're also saying about high trust societies if

00:32:48.980 you are in if you are trying to be a high trust society you are very good at self-policing because

00:32:55.340 you understand that any bad actor within your group may make your group be perceived as an

00:33:00.120 existential threat to others yeah you you know you know how you stop countries from going around and

00:33:04.120 making threats like iran has been doing and north korea does and stuff like that anytime somebody

00:33:08.620 does that you eradicate them yeah but also anytime anyone within your own group does that you

00:33:14.080 eradicate them because if if people within your own country within your own borders or your own

00:33:18.560 culture are doing that everyone's gonna start thinking that your entire group is too dangerous

00:33:24.580 to let a lot of people secretly don't want a high trust world they say they want a high trust world

00:33:29.700 and i'm like okay so you're okay with the constant monitoring that every high trust society ever in

00:33:33.440 human right and the harsh consequences for being a bad actor yeah you're okay with that and they're

00:33:37.560 like oh well i don't want to you know constant monitoring of everything i do in public it's like

00:33:41.480 well that's what high trust is and that's where we're going so buckle up you know i mean that's

00:33:47.680 true that is so true that high trust is extremely close monitoring and scrutiny you're constantly

00:33:53.740 being watched and judged if you live in a high trust society huh and the if we live in a global

00:34:00.600 community we can handle some of the you know you can handle this at the level of an individual but

00:34:05.160 we can live in a high trust collection of countries where we don't have endless arguing

00:34:10.060 in a bureaucracy i mean you basically just recreated the vatican if you've done that

00:34:15.280 with like the u.n or something right like these these always lead to the worst of actors gaining

00:34:21.840 the most power just allow the competent power like where where power is judged by competence

00:34:29.060 which is what we're seeing in the world right now in the world of you know capitalism and stuff like

00:34:32.660 that you can let them decide like people like oh aren't you scared of of palantir i love palantir

00:34:37.280 man palantir is the best people like oh they're in the peter thiel network they love palantir

00:34:42.060 do i love palantir because palantir is what makes our country high trust yeah okay and can make it

00:34:47.880 high trust again flock and all these other monitoring explain how flock works well flock

00:34:54.000 posts cameras like i think say you're a laundromat say you're a random store flock may approach you'd

00:34:59.520 be like hey can i put up a camera i think there's some kind of money exchange like you'll get money

00:35:03.040 from like on a subscription basis or contract basis for having one of their cameras in your area

00:35:07.480 then flock sells as a service to law enforcement agencies the u.s government presumably other

00:35:13.620 parties as well who want the data information on the movement of cars and individuals and it can

00:35:19.440 give you either anonymized or non-anonymized data, depending on who you are. And if you have a

00:35:24.460 warrant on the movements of people, people are really mad about Flock because it's being used

00:35:31.320 by ICE to find people and arrest them. So that's, I think that's primarily why people are extra mad

00:35:37.260 because anything ICE does is like, oh, I have to get super angry. We're enforcing our laws.

00:35:42.800 The bad news for you though, Malcolm is I agree with your argument. I think it's really important,

00:35:47.360 but that is not at all what's being discussed. What's being discussed is like, well, how are

00:35:52.160 the AI companies going to build safeguards to stop this from happening, which is stupid because 0.90

00:35:57.700 literally any tool can be a weapon. Anything that is useful, especially useful enough can become a

00:36:05.380 weapon. The first ever real dead body I ever saw in my life, I was in an Oakland court during like,

00:36:14.220 I don't know it was a lunch break and I was going in as a student like for one of those like see

00:36:19.560 what it's like to be a lawyer days and of course it's Oakland California so there's going to be

00:36:23.600 some violent murder case we're just like walking into for a pizza party in a courtroom and on the

00:36:29.380 evidence board was an image of someone who had been brutally murdered by a hammer and I saw just

00:36:33.780 how much damage a very simple tool that is in most every household can do to someone and I've

00:36:41.220 never looked at hammers the same way it can but it's super like anything that's useful can be a

00:36:46.660 weapon you can't you cannot nerf ai so much that it can't hurt anyone it can hurt with the most

00:36:53.380 toothless version of itself and so stop trying as the technology increases in terms of its capacity

00:36:59.680 for danger yeah our vigilance against bad actors it needs to become more severe and more ruthless

00:37:06.740 Well, here's the irony though. Like here's one person's reaction to the article because a common response to this news on X is people noting the irony and how safeguards worsen the AI experiences for them personally, but not for the terrorists. Like at Walker America on X posted AI frontier models won't tell a Canadian how to resist totalitarian surveillance, but tell a jihadist how to build a bomb. It's almost like they're doing it on purpose.

00:37:32.140 But you see also the irony in his own message for your message, which is that like, here's this person who hates the idea of being monitored, being mad about AI also being nerfed because it's not controlling for the terrorists.

00:37:47.700 Like everyone's talking past your argument and completely missing it and not even aware that it exists, which is kind of, it's frustrating, but it is nowhere in the discourse.

00:37:56.080 I can tell you that. Okay. And the only thing where surveillance turns up in the discourse is

00:38:00.940 I'm mad about surveillance. That's yeah. No, I mean, if, if, if, if we live by the values that

00:38:07.200 we espouse, right. Like long-term, what does techno Puritan communities look like? I would

00:38:13.860 imagine constant AI monitoring of everything you do. Like I would not mind a society in which

00:38:19.680 AI suggests, like it can give you a morality score that you can trust because it's coming

00:38:24.400 from like the town's ai network and then other people and it'll be like oh you'd be like a good

00:38:28.640 match for x person who's also in the town who's also part of the community and that would probably

00:38:33.100 be a better match than anything because you know they're not lying to you about the person you know

00:38:37.500 they see the person day and night and everything like that and they see who they really are right

00:38:42.120 you know or who they're able to fake being 24 7 which is as good as really being a thing right

00:38:47.700 yeah because the moment eight so i i think with with there's a lot of positive externalities to

00:38:53.580 high-trust cultures as well, right? Well, keep in mind, so just, this is also something that

00:38:57.480 not Aldous Huxley alludes me into, but in the EU, I'm just going to link from a tweet he shared

00:39:04.060 with me on X. They basically made this monitoring legal and possible. They call it chat control.

00:39:10.800 So at Level's bio posted, chat control has passed. They can and now will legally scan any person's

00:39:16.940 messages, emails, and photos you send without a warrant. And this is what I'm really against,

00:39:21.860 is stuff that is within the home yeah now if if you're within a techno puritan community and it's

00:39:27.720 a techno puritan ai managed by techno puritan authorities like within our culture that's fine

00:39:31.460 yeah but if it's people outside our culture like a bureaucracy or the european government like the eu

00:39:36.540 which is even worse yeah i mean this is an organization that appears to have the active

00:39:40.500 goal of exterminating you yeah like that is part of their agenda when we have seen in other episodes

00:39:47.040 that they will hide crimes murders grapes when they're committed by migrant groups or minority 1.00

00:39:52.720 populations they want you eradicated yeah and that's why with these groups it's it's important 0.99

00:39:58.580 to note that like you're gonna confuse people though because you're like monitoring is great

00:40:01.960 and then the eu is well when palantir is doing it not when the eu is doing it i don't think people

00:40:08.600 see the difference palantir is culturally aligned with us i know that but like a bunch of tech

00:40:15.680 nerds who want to eradicate terrorists right like yeah now could palantir be eventually

00:40:21.580 ideologically captured yes and if they are then i wouldn't want them monitoring us right but as 0.78

00:40:27.000 the world becomes a more dangerous place the only way that we can maintain and again i think within

00:40:33.480 a society if you're monitoring like what's the ethical way to do it if you can't ensure that

00:40:38.000 your group is the one that's controlling the technology it's only monitor things outside the

00:40:42.440 house i think that that is ethical within any society yeah yeah i think any any public space

00:40:47.100 that yeah any no no not public space anywhere outside the house i think businesses public

00:40:52.280 spaces commercial spaces i think that's why how flock is used is is totally fair game because

00:40:58.400 these businesses are consenting to have the cameras installed in their parking lots which

00:41:02.560 of course now people are vandalizing and destroying but also i see your point about

00:41:05.980 the eu it's it's horrible i even the undemocratic way in which this was passed because this wasn't

00:41:11.500 something that was like oh let's take it slow let's really talk this through no the way they

00:41:16.280 passed this law when the majority of university the european parliament was actually against it

00:41:21.600 as it just so happens is they waited until most of the members were on holiday because as everyone

00:41:28.360 knows europe is freaking hot right now and no one can work because they can't think because they

00:41:32.400 don't use air conditioning and so there were only a few present still and then they created an urgent

00:41:38.160 vote that enabled it to pass through so they they basically just circumvented the democratic process

00:41:44.280 by like probably coordinating with the supporters of the bill being like hey let's all agree

00:41:50.680 to not disappear on vacation like everyone else right now which is like so unusual in europe

00:41:57.300 to to collude to make this vote pass so yeah it's i'm just glad it's europe and not my up

00:42:04.600 it's yeah right i'm glad it's europe yeah man you guys aren't revolting not my problem not my

00:42:11.560 problem right now i do not want to be there right now but yeah this is i'm with you i don't really

00:42:16.940 know how to navigate it because look as much as i agree with your stance i don't even know if it's

00:42:22.340 worth really talking about that much beyond this because it's not going to happen no one's going

00:42:26.940 to systematically take out these groups so what we need to do like in terms of what is within

00:42:32.860 No, as we build the techno-feudal future of humanity, the post-nation-state future, we need to build ideologies that some of these groups can adapt, which can handle the work.

00:42:45.580 Because disproportionately, it's going to be a few of the tech moguls who have a majority of power.

00:42:51.120 And if some of those come out of the techno-Puritan community and they can build techno-Puritan

00:42:55.820 breakaway states or, you know, sovereign cities, stuff like that, then you don't need many

00:43:01.300 of those to keep the world in line.

00:43:04.620 And fortunately, there's already a very large one that's shown willing to play ball, which 0.98

00:43:09.680 is Israel. 0.94

00:43:11.020 So, you know, you just have the wider Israel network that handles this BS. 0.54

00:43:17.580 Yeah. 0.88

00:43:17.880 Okay.

00:43:18.920 That's fair.

00:43:19.720 i'm hoping for the best and also we need to maintain access to non-nerfed ai because

00:43:24.700 one way this could go wait it's real restrains themselves sometimes and i'm like 0.95

00:43:29.160 yeah and the people who are voting for that restraint are they the ones having kids

00:43:32.900 no if you look at somebody's talking about israeli friend of ours like if you look at

00:43:37.620 far left versus far right protests in israel he goes the leftist ones like everyone is 70 or over

00:43:42.660 they're like it's really weird and they're like in the younger ones it's like all young people

00:43:46.220 and everything like that so yeah yeah in in terms of what the future is going to look like in a in

00:43:52.840 a positive future and again this is why i think when i think about like the the sort of like

00:44:00.560 techno future that i'm looking forward to that's why i always go as a terran empire i think fixing

00:44:06.740 a lot of the world problems today is going to look like what a lot of people get scared about

00:44:12.340 when they're when they're looking at like mirror world star trek we never did our episode on that

00:44:16.540 being like that's but also yeah as people know everyone had like a mustache right yeah and all

00:44:21.820 the girls dressed extra sexy and everyone's mad that like the right today is like what are you

00:44:26.920 guys with like cat girls and like elon has like a sexy grok and you guys have like not safe for

00:44:32.500 work stuff and i'm like well our version of christianity is just significantly more accepting

00:44:35.960 of not safe for work stuff and like using hotness as a trap for our enemies or to make ideas

00:44:42.320 more appealing right like hacking the human brain to achieve our ends is completely one of the cards 0.77

00:44:48.680 on the table for us and so that means we will have the terran empire with the ai monitoring everyone 1.00

00:44:56.900 strict order in society but also all the women in miniskirts you know we've got to have oh nothing 0.95

00:45:03.400 makes me hornier than torturing someone i'm horny all the time and being horny always gets me in the 0.59

00:45:09.060 mood to torture on you can lose a whole day to that cycle live in the dream yeah well while 0.92

00:45:15.100 they're still hot you know when they get older we go trad maxing right like trad maxing is the

00:45:20.040 style that like not that someone couldn't still be you'd be gross if you were many malcolm the 0.88

00:45:24.360 word is hag maxing trad maxing is when you dress in a maliciously traditional style which is what 0.94

00:45:33.380 you do what i should do more the only reason i'm not doing it today that's just too hot to wear

00:45:36.800 anything more than a t-shirt sorry about that guys i couldn't bear it yeah here i am wearing

00:45:40.980 however many layers and i'm the one on a treadmill desk all day you gotta just get with the program 0.70

00:45:46.360 malcolm pain pain is what devotion is old i fucking love you you're awesome no but i i love 0.70

00:45:55.300 the idea of these like tech sexy like very high trust is the word we're gonna use for these 0.93

00:46:02.760 societies with those you know high trust banners hanging everywhere and those you know a high trust

00:46:09.020 way of walking high trust walking oh is this i guess consistent signatured walking like in

00:46:16.220 inquiry doctor's book little brother i'll put in a short clip of some high trust walking here

00:46:21.320 it's gonna be through stepping oh god welcome in actually walking was a big theme in cory

00:46:32.640 doctor's book little brother in that like of course cameras do gate recognition so the way

00:46:37.960 that the protagonist avoided gate recognition was to put gravel in his shoes so that he would walk

00:46:44.760 funny which is not a bad idea presumably that might work so i don't know whatever high high

00:46:52.640 i think you could catch somebody doing that pretty quickly as soon as you learn that that trick

00:46:57.180 yeah this is our high trust society is going to be so high trust people are going to be what i'm

00:47:02.240 doing simone is i'm using the term high trust as a euphemism for another word that scares it might

00:47:09.280 oh god i am way too autistic for you so we'll have our high trust nation states that will

00:47:17.280 be run by our tech lords and i'm okay with that yeah if elon was like literally dictator of like

00:47:24.020 my society i'd be totally chill with it i'd be like yeah sure whatever you want i'd be like you

00:47:27.840 probably thought even when i disagree with him i'd be like you probably thought this through

00:47:31.280 more than i have right like honestly if peter teal was in charge of our society i would be

00:47:35.400 totally chill. He's a thoughtful dude. Yeah. If Chamath was in charge, I'd be down with it.

00:47:40.040 Oh yeah. Yeah. Yeah. Most of the tech lords, JD Vance, I'd be down with it. You know, this,

00:47:45.580 this whole democracy thing had worked for a while. It worked for a while, but the problem is,

00:47:50.580 is that letting average people make a decision turns out to have its limits.

00:47:56.800 And the founding fathers knew this, like they tried to protect us from this.

00:48:00.980 yeah they they did not recommend democracy at all not even close to it well because

00:48:08.380 democracy like intrinsically leads to lower trust societies over time yeah no their system was just

00:48:15.340 literally a vote for an educated person that you trust and just let them handle it okay don't get

00:48:23.940 involved and we just failed to execute on that which is our bad but whatever anyway i love you

00:48:32.300 very much and i appreciate you having this conversation with me i know you're like yeah

00:48:36.520 i have to work but i want to talk with you i love you you're pretty well i've i've been hopefully

00:48:42.760 getting some stuff done in the background see when i talk you can work and when you talk i can

00:48:48.140 scroll well no even when you when i talk i'm scrolling instagram because i can't look at my

00:48:52.480 face but that's okay because instead i pick up weird stuff from instagram i love you malcolm

00:49:00.100 and we'll we'll hopefully not die as terrorists figure out how to target our dna and release

00:49:07.300 goodness knows what to just kill us as we randomly go about our lives bye bye

00:49:17.060 Okay, let me come and look.

00:49:23.420 We got so much.

00:49:25.940 And a big bummer.

00:49:26.960 Whoa. 0.97

00:49:27.920 Look at that big bull.

00:49:29.540 And all of those.

00:49:30.680 Impressive.

00:49:32.660 And that.

00:49:34.480 Yeah?

00:49:35.960 And did you catch him in that net?

00:49:42.160 Trying to get it out, Titan?

00:49:45.700 You got beeping?

00:49:47.060 Oh, put it back. Is that a tadpole?

00:49:52.760 Yeah, it's doing a swipe cycle of turning into a frog.

00:49:57.680 Yeah, don't tip it over, Andy.

00:50:01.700 No, Andy, we just don't want them to go disappearing, girl. It's okay.

00:50:09.280 It's okay. Hey.

00:50:13.660 I am very impressed with the tadpole.

00:50:17.060 I don't think I've ever seen one at that stage of the life cycle before.

00:50:22.100 Wow.

00:50:24.680 Daddy, give me my book.

Based Camp - July 15, 2026

How AI Revolutionized Terrorism

Episode Stats

Summary

Transcript