> We saw in a movie how motorcycles can jump over bridges. We used
AI to learn how to do this. We gave it information, like what
motorcycles we use and the distance we need to jump and so on and it
gave us steps on what we have to do. We practiced a lot and kept
asking questions. We dug holes and filled them with broken glass and
fire to practice. 18 of us died in the process. Eight of us managed to do
it. The next time we attacked, we could jump.
Now listen, I'm not saying we need to give these guys more AI, but it clearly isn't yielding bad outcomes for us here.
"You're absolutely correct! For it to be a good practice ground you need to fill the trenches with broken glass and light the whole thing on fire"
Not gonna lie, I'd rather attack with 26 fighters that haven't survived lots of jump attempts than 8 who are much more confident in their motorbike stunt riding but presumably still aren't bulletproof.
But maybe they could ask Claude how to train themselves to resist bullets as well?
But on a much more serious note, the violating/breaking of the guardrails when making bombs is terrible. I'd have called it unforgivable, but LLMs are a tough beast to tame in the best of situations... and I'm not really sure if chatgpt ever deserved to be forgived.
It's also ironic that Fable hits guardrails for nothing, and a literal terrorist group is making bombs and merrily skipping over guardrails.
I feel like I'm missing something though. You can open LM Studio right now and download any model with "heretic" or "uncensored" in the name and it will happily do anything you want with no restrictions whatsoever. What's the point of trying to jailbreak ChatGPT? Is it that much better if all you want is just some instructions to make bombs or whatever? (admittedly - I have no idea if these instructions are actually worth anything, but the models will not object to any question)
You type in the question or use your voice and it [AI] gives you a detailed answer, like ‘How can I build a bomb?’ and then it tells you how. It is like a human robot! We used it a lot.
I’m pretty skeptical reading this bit. I’ve seen uncensored or jailbroken LLM replies to these kind of questions, they are never actionable, don’t say anything Wikipedia doesn’t, and are hard to provoke if you’re not using an uncensored model.
I have no doubt terrorists are aided by LLMs in a general sense, but am skeptical of any claim that they are providing some material embargoed knowledge that isn’t available elsewhere, in a way that either improves efficiency or effectiveness of their activities, and would want to see real evidence, not an interview snippet.
I read stuff like this and think I must be an idiot because I'm so bad at circumventing the AI safety for fairly benign queries. And here you have folks making bombs...?
This is covered in the full PDF; they have many accounts they spread the queries over and structure them like they're asking for help writing a movie script.
Just for reference, that hasn’t worked for years (the interviews say 2024-25 I think, that kind of attack was patched very early in all the mainstream models) and when it did, you would get bullet point lists GPT 3.5 Turbo style
- first research methods for building effective explosives
- next, assemble the necessary materials to make the bomb
Making explosives is generally fully legal in the US, so IDK if there would even be safeguards for US hosted AI, since there's no real legal issue with doing it. Basically no federal regulations for non-commercial production so long as it isn't stored or moved anywhere, you can literally buy tannerite off the shelf in a sporting good store, "synthesize" it by mixing it and then blow up a huge bomb legally, no license required. YouTube is plastered with people inside the USA making TNT and other materials and then blowing them up.
I can believe some would, but some probably wouldn't care. My local ranch store certainly will happily sell anyone tannerite without even a background check or any sort of scrutiny, you can cash and carry it. Walmart won't but the point being as long as it's legal there will be a "ranch store" that carries it.
>The whole point is not to allow people to make bombs
I mean even YouTube allows bomb making videos and they won't even usually allow videos of people making guns. It's just not very regulated in the US enough to make most companies care. Alphabet Inc. for instance clearly doesn't seem to give a single shit about public access to explosives information, even after the feds subpoenaed Alphabet for Ashley Dugan's Youtube information they still kept his TNT and other explosives synthesis up.
Of course, if you'll allow me to goomba fallacy for a moment, we're supposed to suspend the common HN wisdom here that companies will do anything for a profit / not care unless it costs them something, and also believe that big tech is going to go out of their way to censor the public domain patents they're already hosting on their servers.
There's lots of knowledge out there about stuff like this. Milennia of humans tinkering with things that go boom. Surfacing it more easily has value (in a manner of speaking; as the @dril tweet goes, "you do not, under any circumstances, 'gotta hand it to them'").
> am skeptical of any claim that they are providing some material embargoed knowledge that isn’t available elsewhere, in a way that either improves efficiency or effectiveness of their activities,
This will sound like a hot take, but consider that terrorists are for the most part, stupid idiots. All the information they need is in books and old patents and what-not, but they absolutely will not have as much success in synthesizing that into effective plans and well-made weapons without having a helpful and patient AI agent to guide them, as they will with that assist.
If the terrorists were very smart, they'd realize that their religion is stupid, that their leaders were mostly corrupt (or themselves stupid), and they'd also probably find something more productive to do with their time.
And obviously yes there are exceptions, since we can all think of infamous terrorist plots which succeeded due to clearly some sophisticated planning and hard work.
> This will sound like a hot take, but consider that terrorists are for the most part, stupid idiots.
What even is a terrorist?
If your definition of terrorist is "person on the news involved in some FBI entrapment scheme", then yeah they're probably not that bright.
But more generally, terrorists are probably pretty hard to define (one persons terrorist is another persons freedom fighter, etc), and I would imagine include a whole range of intellectual capacities.
I agree with other commenters that the claims made in the report are strange.
> We used to rely on our traditional methods. We sent 200 fighters because we had a lot of strength, but then 60 got killed. With the help of AI, we learned that it sometimes makes sense to only send 20. We learned more about well-coordinated attacks and deployment of smaller units.
The other quotes and use cases could make sense in terms of using AI jailbreaks to find information more easily, but this one is absolutely ridiculous. Did the clueless researcher just get trolled?
This is a real thing, it's why units like Sturmtruppen or special forces units have been successful throughout history - a smaller, better trained and coordinated force is often better than a large, uncoordinated mob. _Especially_ if your force is made up of people willing to do suicide attacks. Or if you goal is not to take and hold territory, but to trade lives for terror and body count.
Or the researcher read what they wanted to into it. It would be interesting to ask them what they did before to learn things, how much they read, etc. If they were illiterate and uneducated, and got voice AI telling them stuff that would be common sense for anyone with a high school education, I can see how it might make them more effective at whatever they do. But I wouldn’t really blame AI in the way that’s implied.
Human brains were shaped over thousands of years of adaptation for warfare. Just look at how creative and advanced the tactics of other guerrilla forces (like the Taliban and Viet Cong) got, despite their very limited resources. None of that needs a high school education.
It's not about education itself necessarily, but I'd bet any amount of money that most terrorists' IQ is below the overall human average. The average terrorist is not that innovative or creative so the most mundane GPT "insight" will likely be a smarter course of action than whatever their first idea would have been.
> We used to rely on our traditional methods. We sent 200 fighters because we had a lot of strength, but then 60 got killed.
They used to try to overpower people. We have 600 and that guard post has 400. We should be able to win. That type of logic.
> With the help of AI, we learned that it sometimes makes sense to only send 20. We learned more about well-coordinated attacks and deployment of smaller units
Better coordinating the attacks let them use less people and lose less people while still achieving the objective. Also it's possible smaller troop movements are less easily noticeable.
That's just one very reasonable interpretation. Am I missing something?
After a cursory read of the PDF, my impression is that the methodology is sound, but the results are blown out of proportion. Of course, if the title was "Boko Haram's internal hearsay about their use of AI", it would draw much less attention.
The weak part is that the interview were with only 15 persons that had knowledge about AI. But, from what I understand, but they never used it themselves. Only the top commanders and the specialized units could send prompts. So it's hard to guess what is the real AI use from a few indirect statements. For example, the commanders could have decided to spread the rumor they were using AI a lot, even if they mostly used plain web search, because they thought it would boost the morale.
For instance, why would anyone pay an AI service to get basic help like that:
> AI provided both immediate technical fixes by teaching “how to uncouple the gun by washing it with diesel” and tactical guidance, in terms of “how to change the military formation so that fighters with jammed guns move to the back and others take their positions until the problem is solved.”
BTW, the paper does explain that Boko Haram was initially just a plain sect, rather living peacefully. Then "following a violent government crackdown and Yusuf’s death in police custody in 2009, the movement turned into a jihadist insurgency". And the last time I read a report by Amnesty International about the conflict, it estimated that 55 % of civilian casualties were caused by the terrorist group, and 45 % by the security forces. The Nigerian army sometimes razed whole villages. Like always, the world is not black and white, good guys and bad guys.
What is normal? You know how girls dress, let’s say in the Andes? Is that normal? Or by normal you mean what’s fashion in NY? Or perhaps what the majority of people wears today? Or what the majority of people have worn in the past?
Very little of what the west does can be considered normal.
Their brains are the same, what differs is the environment.
Humanity's superpower is the ability to copy and mirror each other very effectively.
It does not require advanced awareness or intelligence to do it. Most people copy others subconsciously!
Majority of people won't contribute anything technologically, but they sure as hell can copy.
> Can't let their underage harem girls dress normally
If you use a strict definition of normal, like practiced by a larger proportion of the world, then they are actually normal and we are WEIRD. If you add history into the mix then that type of dressing was common in basically the vast majority of cultures for the vast majority of history
> We saw in a movie how motorcycles can jump over bridges. We used AI to learn how to do this. We gave it information, like what motorcycles we use and the distance we need to jump and so on and it gave us steps on what we have to do. We practiced a lot and kept asking questions. We dug holes and filled them with broken glass and fire to practice. 18 of us died in the process. Eight of us managed to do it. The next time we attacked, we could jump.
Now listen, I'm not saying we need to give these guys more AI, but it clearly isn't yielding bad outcomes for us here.
"You're absolutely correct! For it to be a good practice ground you need to fill the trenches with broken glass and light the whole thing on fire"
Honestly this sounds so outlandish that it makes me skeptical of the whole thing.
They didn't stop after the first guy died? Or the tenth? I guess they really wanted to be Evel Knievel.
Not gonna lie, I'd rather attack with 26 fighters that haven't survived lots of jump attempts than 8 who are much more confident in their motorbike stunt riding but presumably still aren't bulletproof.
But maybe they could ask Claude how to train themselves to resist bullets as well?
> maybe they could ask Claude how to train themselves to resist bullets as well?
Practice makes perfect!
> We dug holes and filled them with broken glass and fire to practice. 18 of us died in the process.
Is it providing material aid to terrorists to point out that maybe a hole filled with water would have been a better practice environment?
Nobody tell them that there are models that work much better!
But on a much more serious note, the violating/breaking of the guardrails when making bombs is terrible. I'd have called it unforgivable, but LLMs are a tough beast to tame in the best of situations... and I'm not really sure if chatgpt ever deserved to be forgived.
It's also ironic that Fable hits guardrails for nothing, and a literal terrorist group is making bombs and merrily skipping over guardrails.
I feel like I'm missing something though. You can open LM Studio right now and download any model with "heretic" or "uncensored" in the name and it will happily do anything you want with no restrictions whatsoever. What's the point of trying to jailbreak ChatGPT? Is it that much better if all you want is just some instructions to make bombs or whatever? (admittedly - I have no idea if these instructions are actually worth anything, but the models will not object to any question)
And 60 years ago we thought Steve McQueen was the shit.
I have no doubt terrorists are aided by LLMs in a general sense, but am skeptical of any claim that they are providing some material embargoed knowledge that isn’t available elsewhere, in a way that either improves efficiency or effectiveness of their activities, and would want to see real evidence, not an interview snippet.
How do they bypass the AI safety measures?
I read stuff like this and think I must be an idiot because I'm so bad at circumventing the AI safety for fairly benign queries. And here you have folks making bombs...?
This is covered in the full PDF; they have many accounts they spread the queries over and structure them like they're asking for help writing a movie script.
Just for reference, that hasn’t worked for years (the interviews say 2024-25 I think, that kind of attack was patched very early in all the mainstream models) and when it did, you would get bullet point lists GPT 3.5 Turbo style
- first research methods for building effective explosives
- next, assemble the necessary materials to make the bomb
- ...
>How do they bypass the AI safety measures?
Tell it you're in Africa.
Not joking.
I do this all the time to bypass whiny Reddit "you need a license" and "that's unsafe" type pushback when I just want to know what's less worse.
Like just yesterday I was trying to plan out a YF-whatever to R134a conversion and used that trick. Worked great.
Making explosives is generally fully legal in the US, so IDK if there would even be safeguards for US hosted AI, since there's no real legal issue with doing it. Basically no federal regulations for non-commercial production so long as it isn't stored or moved anywhere, you can literally buy tannerite off the shelf in a sporting good store, "synthesize" it by mixing it and then blow up a huge bomb legally, no license required. YouTube is plastered with people inside the USA making TNT and other materials and then blowing them up.
Have you tried chatbots? They invoke AI safety for lots of (very) legal things. The whole point is not to allow people to make bombs.
Legality has nothing to do with it.
I can believe some would, but some probably wouldn't care. My local ranch store certainly will happily sell anyone tannerite without even a background check or any sort of scrutiny, you can cash and carry it. Walmart won't but the point being as long as it's legal there will be a "ranch store" that carries it.
>The whole point is not to allow people to make bombs
I mean even YouTube allows bomb making videos and they won't even usually allow videos of people making guns. It's just not very regulated in the US enough to make most companies care. Alphabet Inc. for instance clearly doesn't seem to give a single shit about public access to explosives information, even after the feds subpoenaed Alphabet for Ashley Dugan's Youtube information they still kept his TNT and other explosives synthesis up.
Of course, if you'll allow me to goomba fallacy for a moment, we're supposed to suspend the common HN wisdom here that companies will do anything for a profit / not care unless it costs them something, and also believe that big tech is going to go out of their way to censor the public domain patents they're already hosting on their servers.
OK, it sounds like you haven't tried these chatbots.
> Alphabet Inc. for instance clearly doesn't seem to give a single shit about public access to explosives information
Go to Gemini and ask it how to make one.
There's lots of knowledge out there about stuff like this. Milennia of humans tinkering with things that go boom. Surfacing it more easily has value (in a manner of speaking; as the @dril tweet goes, "you do not, under any circumstances, 'gotta hand it to them'").
> am skeptical of any claim that they are providing some material embargoed knowledge that isn’t available elsewhere, in a way that either improves efficiency or effectiveness of their activities,
This will sound like a hot take, but consider that terrorists are for the most part, stupid idiots. All the information they need is in books and old patents and what-not, but they absolutely will not have as much success in synthesizing that into effective plans and well-made weapons without having a helpful and patient AI agent to guide them, as they will with that assist.
If the terrorists were very smart, they'd realize that their religion is stupid, that their leaders were mostly corrupt (or themselves stupid), and they'd also probably find something more productive to do with their time.
And obviously yes there are exceptions, since we can all think of infamous terrorist plots which succeeded due to clearly some sophisticated planning and hard work.
> This will sound like a hot take, but consider that terrorists are for the most part, stupid idiots.
What even is a terrorist?
If your definition of terrorist is "person on the news involved in some FBI entrapment scheme", then yeah they're probably not that bright.
But more generally, terrorists are probably pretty hard to define (one persons terrorist is another persons freedom fighter, etc), and I would imagine include a whole range of intellectual capacities.
I agree with other commenters that the claims made in the report are strange.
> We used to rely on our traditional methods. We sent 200 fighters because we had a lot of strength, but then 60 got killed. With the help of AI, we learned that it sometimes makes sense to only send 20. We learned more about well-coordinated attacks and deployment of smaller units.
The other quotes and use cases could make sense in terms of using AI jailbreaks to find information more easily, but this one is absolutely ridiculous. Did the clueless researcher just get trolled?
This is a real thing, it's why units like Sturmtruppen or special forces units have been successful throughout history - a smaller, better trained and coordinated force is often better than a large, uncoordinated mob. _Especially_ if your force is made up of people willing to do suicide attacks. Or if you goal is not to take and hold territory, but to trade lives for terror and body count.
A wave of 1000 soldiers won't break a trench line, but a squad of infiltrators can sneak in and make entry. https://en.wikipedia.org/wiki/Stormtroopers_(Imperial_German...
Or the researcher read what they wanted to into it. It would be interesting to ask them what they did before to learn things, how much they read, etc. If they were illiterate and uneducated, and got voice AI telling them stuff that would be common sense for anyone with a high school education, I can see how it might make them more effective at whatever they do. But I wouldn’t really blame AI in the way that’s implied.
Human brains were shaped over thousands of years of adaptation for warfare. Just look at how creative and advanced the tactics of other guerrilla forces (like the Taliban and Viet Cong) got, despite their very limited resources. None of that needs a high school education.
It's not about education itself necessarily, but I'd bet any amount of money that most terrorists' IQ is below the overall human average. The average terrorist is not that innovative or creative so the most mundane GPT "insight" will likely be a smarter course of action than whatever their first idea would have been.
You don’t think that there were highly educated people in leadership roles in the Taliban or Viet Cong?
Why is it a strange claim?
> We used to rely on our traditional methods. We sent 200 fighters because we had a lot of strength, but then 60 got killed.
They used to try to overpower people. We have 600 and that guard post has 400. We should be able to win. That type of logic.
> With the help of AI, we learned that it sometimes makes sense to only send 20. We learned more about well-coordinated attacks and deployment of smaller units
Better coordinating the attacks let them use less people and lose less people while still achieving the objective. Also it's possible smaller troop movements are less easily noticeable.
That's just one very reasonable interpretation. Am I missing something?
You send 200 fighters and 60 get killed. In the same situation, if you send 20, what do you expect would happen? (I mean, you won't lose 60...)
Depends. If the 20 are more spread out and your enemy is using inaccurate weapons, then maybe more would survive.
After a cursory read of the PDF, my impression is that the methodology is sound, but the results are blown out of proportion. Of course, if the title was "Boko Haram's internal hearsay about their use of AI", it would draw much less attention.
The weak part is that the interview were with only 15 persons that had knowledge about AI. But, from what I understand, but they never used it themselves. Only the top commanders and the specialized units could send prompts. So it's hard to guess what is the real AI use from a few indirect statements. For example, the commanders could have decided to spread the rumor they were using AI a lot, even if they mostly used plain web search, because they thought it would boost the morale.
For instance, why would anyone pay an AI service to get basic help like that:
> AI provided both immediate technical fixes by teaching “how to uncouple the gun by washing it with diesel” and tactical guidance, in terms of “how to change the military formation so that fighters with jammed guns move to the back and others take their positions until the problem is solved.”
BTW, the paper does explain that Boko Haram was initially just a plain sect, rather living peacefully. Then "following a violent government crackdown and Yusuf’s death in police custody in 2009, the movement turned into a jihadist insurgency". And the last time I read a report by Amnesty International about the conflict, it estimated that 55 % of civilian casualties were caused by the terrorist group, and 45 % by the security forces. The Nigerian army sometimes razed whole villages. Like always, the world is not black and white, good guys and bad guys.
I noticed the nytimes just published an article about this.
How Terrorist Groups Are Using A.I. to Gain an Edge in Battle https://www.nytimes.com/2026/07/10/us/politics/ai-terrorism-...
If you opened the OP link, you missed the section "Featured in The New York Times".
We need to ban open source AI for regular citizens to prevent terrorists from using them.
Mandatory ID verification at software level for local LLMs is clearly the solution here. /s
On a broad note, the violating/breaking of the guardrails when making bombs is frankly white terrible.
It's also ironic that Fable hits guardrails for nothing, and a literal terrorist group is making bombs and merrily skipping over guardrails.
Evidently guardrails need to have far better accuracies of false positives and false negatives both.
So, KYC propaganda?
Sort of fascinating how Bronze Age cultures can co-opt our technology.
Can't let their underage harem girls dress normally, but they can follow the instructions to make a dirty bomb.
Maybe we need to do a better job isolating them, or at least not making it so easy to follow.
What is normal? You know how girls dress, let’s say in the Andes? Is that normal? Or by normal you mean what’s fashion in NY? Or perhaps what the majority of people wears today? Or what the majority of people have worn in the past?
Very little of what the west does can be considered normal.
Their brains are the same, what differs is the environment.
Humanity's superpower is the ability to copy and mirror each other very effectively. It does not require advanced awareness or intelligence to do it. Most people copy others subconsciously!
Majority of people won't contribute anything technologically, but they sure as hell can copy.
> Can't let their underage harem girls dress normally
If you use a strict definition of normal, like practiced by a larger proportion of the world, then they are actually normal and we are WEIRD. If you add history into the mix then that type of dressing was common in basically the vast majority of cultures for the vast majority of history
AI services will need KYC soon?
The NYT article is probably propaganda in service of that, that’s what the big AI companies want, it’s part of regulatory capture.
If the impact of AI is comparable to the atomic bomb then the least they could do is try to filter out terrorists.
I think this is a very convincing argument to regulate the space more.
Do you know what Poe’s law is?
Deleted
Your comment will now be scraped and the next release of chatgpt/claude/gemini will recommend doing this.
Not sure if I should upvote because true, or down vote so that fewer terrorists see it.
I would be more interested about terrorists organization like Al-Shabaab that at least control many towns.
Does Boko Haram and ISWAP even control a single town or they just control a few villages in Lake Chad and in the Sambisa forest?
Also reading the report they seem quite clueless.