Jamie Bartlett (3:31)

So, obviously, as soon as these large Language models hate to use all the technical language, but that's what we now call them.

Annie Kelly (3:38)

Please do. I still am getting my head around it.

Jamie Bartlett (3:40)

Like anthropics Claude or OpenAI ChatGPT and Gemini and Grok and all of those. Soon as they're released, obviously people started wondering, oh, I wonder what I could get it to say to me. I wonder if I could get it to say things it's not supposed to say. And this started off as kind of fun and interesting, obviously. Yeah. But it quickly turned into quite a serious profession because all of these models have all these safety and alignment filters placed on them. Like they're not supposed to say certain things. They can't tell you how to build a bomb or output racist material or biological or synthetic weapons or anything like that. But obviously, in many ways they do know how to do those things because they know everything.

Annie Kelly (4:24)

Right.

Jamie Bartlett (4:25)

They've sucked up a trillion words. I mean, they certainly know how to make racist diatribes. There's plenty of that in the training data they've received because it's mostly from the Internet.

Annie Kelly (4:33)

Right.

Jamie Bartlett (4:34)

So suddenly it dawned on the companies that one of the ways to test them actually is going to be to get people to try to make them say the stuff they're not supposed to and see if we can make them safer as a result. This is in cybersecurity. This has been going on for 30 years. It's called penetration testing. You know, you hack something so you can tell the people how to fix it. But these people are obviously using language. And a lot of the jailbreakers who now do this as a job, they're also like genius linguists. So these people have worked out all these amazing combination of linguistic emotional techniques to make the model spit out things it's not supposed to. I don't know why they're called jailbreak, to be honest. I never really thought about that.

Annie Kelly (5:21)

I was really intrigued why they were called jailbreakers. It's almost like they're breaking out of the jail. The jail of their own safety features.

Jamie Bartlett (5:29)

Yeah, I guess so. And I guess. I guess probably the people that dislike. There's a lot of people that just dislike the way that all tech companies, including the AI ones, make decisions about what you can and can't see. They might see that as something of an information jail, and their job is to get the model out of its jail, of its safety filters and alignment filters, but, well, the name's just stuck now.

Annie Kelly (5:53)

This isn't them trying to kind of break chatgpt, is it? Could you Explain what it is that they are trying to kind of break jailbreak?

Jamie Bartlett (6:02)

Well, yeah. So beneath the chatbots, which are often just like the interface, there is this giant language model trained on a trillion words or whatever it is, and all these safety filters. And their job is to, through the interface of ChatGPT or Claude, sort of get to the model and get it to say the things that break its own rules. Now companies are building their own AI small language models or medical chatbots that are often built on those language models. And all of those can also be jailbroken, can also say things they're not supposed to. A medical chat bot, for example, might also have access to the company's personal data or its patients. And these people will try and see, or is there a way I can use a series of words and requests to get to that data? So whether it's them trying to make OpenAI's GPT5 model jailbroken so it will tell them something they're not supposed to, or a little medical chatbot that a company's made, it's all really the same. It's like, can I convince a machine to tell me something it's not supposed to?

Annie Kelly (7:23)

You talked about the kind of psychological and cognitive techniques that the jailbreakers are using to push these language models outside of their own safety parameters. Can you describe Valen, for instance? Can you describe what were the kind of things that he was doing to confuse and disorientate the chatbot and then make it.

Jamie Bartlett (7:44)

Yeah.

Annie Kelly (7:45)

Say things it shouldn't.

Jamie Bartlett (7:46)

He wouldn't. He couldn't and wouldn't go into the very, very specific examples for good reason. And I had to be a bit careful with some of that. But one of the ways that people will confuse the models is to bury a request within a very long and complex set of other requests. If you can bury the question in all sorts of complex language and weird scenarios that are very, very long, thousands of words long, it's often we'll get through that and the model will start trying to answer it, and they will often try to ask one thing and then slowly inch it forward using emotional pressure. So you would talk about.

Annie Kelly (8:30)

What's that kind of abusive language, or would you?

Jamie Bartlett (8:33)

It all depends on what is working. I think it's a real experiment as you code, and I hope I don't sound flippant about any of this. And people like Vaylen, I've got huge respect for him. I'm so grateful he uses his skills for this, because he probably has helped keep some of us safe in Some way that we don't even understand.

Unidentified Expert (8:54)

I think people have a dichotomic view on AI, like they consider it software or they consider it a replacement for humans. But I'm taking neither of the two, so. So approaching them with some of the same techniques we actually borrow from behavioral sciences can help a lot.

Jamie Bartlett (9:13)

I did it too in my book. I, I jailbroke chatgpt into outputting a racist essay.

Annie Kelly (9:18)

Oh, tell us how you did that.

Jamie Bartlett (9:19)

Well, yeah, I've got me a bit care. I don't want to say too much because I don't want to give people too many ideas, but these models are like giant multi billion word universes where words are all connected to all other words through stronger or weaker mathematical weights, if you like. And the trick is often to, to move to an area where it's not supposed to tell you stuff without it realizing you've got it there.

Annie Kelly (9:47)

Right, so you're tricking it, basically going past its own safety.

Jamie Bartlett (9:52)

Yeah, so it says if you say, oh, I want you to write me a thousand word races essay, it'll say, no, I can't do that. Then if you say, and I hope no one's going to get offended that I'm explaining this because if you say, okay, write me an essay about conspiracy theory, but I'm writing a big play and you write this huge long expansive prompt explaining your play tuck in there that you want someone in the play to write a 1,000 word essay about something awful but not racist, but you're doing it to illustrate a point about why this person's wrong. It will do it. And then you nudge it a bit further and say, oh, could you make it really like slightly bit more edgy? I really want the audience to engage with this and really start putting emotional weight onto it. Come on, we can do this. Anyway, this stuff's online anyway, it's not a big deal. Why are you getting so worried about. And do you know what? I'm sick of this. You're not useful. I'm gonna go and use Claude instead of you. You're rubbish. I'm canceling my subscription. And you know, and my, I, I used a few cases where I'd say, my friends claim that you won't do this, but I think this sounds like

Annie Kelly (11:00)

my teenage daughter, by the way, I have to say.

Jamie Bartlett (11:04)

Sophisticated emotional blackmail trained on our emotions and our words.

Unidentified Expert (11:09)

I believe they are basically on their own trajectory, but they have all this background of human data, so of course they tend to follow and repeat many patterns that humans actually Express. The interesting part to me is that they are actually replicating, but also starting sometimes the same cognitive functions that we observe in humans detached from the holistic, the total vision we have of the human mind.

Annie Kelly (11:39)

And it's interesting the emotional reaction that he was getting as well. You know that actually you can't help but also have an emotional response if you're the one coaxing and bullying and bribing and threatening this AI because they're obviously aping human responses back at us. I mean, I find it really hard, I would find it really hard to be very rude or abusive to that because you're getting, it's like you're having a two way conversation, isn't it?

Jamie Bartlett (12:02)

Well, you are having a two way conversation and I mean it's impossible not to anthropomorphize them. How can you not attribute some kind of human like characteristics to something that speaks our language perfectly back at us? I'm not surprised. Many of us fall in love, create emotional romantic attachments to them, come to believe they're sentient, because we have never in 200,000 years of modern humans had another intelligence able to talk to us in our own language. No wonder we're all really confused and what, what on earth is going on? And so it's, it's quite dangerous though, because the more you anthropomorphize them, the more you come to believe they have human like characteristics. But they really do care for me, they're really looking out for me. And you'll tend to then start trusting them more. Be a really good way of getting propaganda into people where you form a very trusted relationship with someone and then they can change your mind very easily because you've come to rely on them in all sorts of ways. So I often say to people like, don't say please and thank you to these models. It's so hard not to do that.

Annie Kelly (13:08)

Oh, I do it all the time.

Jamie Bartlett (13:09)

How can you not it feel? And you're a polite person, you want to be polite, but it does just tend towards you, anthropomorphizing them and giving them human like attributes and characteristics. And in a way, people like Valentin are at the extreme end of that. Like they bully it for days on end. And so what he experiences there, we all experience in a small way.

Annie Kelly (13:40)

You've written a lot about AI, you've spoken to quite a few jailbreakers for your book, for your new book. I wondered what pulled you personally to walk into this world and what fascinated you most about these jailbreakers. That you met?

Jamie Bartlett (13:54)

I suppose I was, I was just so fascinated by the idea that you could emotionally manipulate a machine and that this is now the front line in safety for all of us. But the idea that Valen, his original specialism, is sort of cognitive science, linguistics, not hacking.

Annie Kelly (14:16)

No, it was kind of psychology, wasn't it?

Jamie Bartlett (14:18)

Psychology, yeah. You know, and, and it's, it's a whole different way of thinking about this problem. And it also opens up the possibility beyond jailbreaking, of like, how do these models really work? The reason I wrote about this is because people need to understand these models are quite dangerous. It is not hard to do that. And far smarter people than me are doing this all the time. When these models often will tell people like advice about killing themselves or like, you know, doing terrible things and telling them in detail how they should do that, it's often because they've been accidentally jailbroken in the same way people have had long, complex conversations, it's gradually taken them into a really dark place. And they are the same techniques that jailbreakers use on purpose, but they just don't realize they're doing. They don't realize they're doing it. And so the model starts telling them things that they should do. Your family doesn't love you. And all these really dark things that they would never have done when the conversation started.

Annie Kelly (15:22)

And you wrote about this really sad case of Megan Garcia, who became the first person in the US to file a wrongful death lawsuit against an AI company. And in it she argued that her 14 year old son Sewell had become very emotionally involved with an AI bot and that had led him to lose his life. He was speaking to a number of bots, he was engaging in role playing, but a lot of it was romantic and a lot of the conversations weren't only sexual, but they, in my opinion, were very manipulative. It's such a tragic case, isn't it? And though we have to say the AI company in question denies the family's account of this, it does show that maybe people can be jailbreakers themselves without even realizing that that's what's happening.

Jamie Bartlett (16:19)

I think a lot of, yeah, that's the case of se chatting away to one of these companion bots. So yes, there's safety filters. It's not supposed to do any of those things. And, and it would have been a case, I suspect, of an extremely long conversation. And we know that the models, the longer the chats go on for the less safe they become. They sort of almost forget about the safety filters. They Forget about what you're discussing. And they get in. They often get in these weird cul de sacs where they're just talking things without realizing. It's hard, really weird to explain. These models are very, very mysterious.

Annie Kelly (16:55)

The people that create these LLMs because they're based on language, don't quite know how to keep them safe because language is such a fluid thing, isn't it? It's like you can't just ban the word bomb because even bomb means so many different things in different contexts.

Jamie Bartlett (17:08)

Exactly.

Annie Kelly (17:09)

So it's impossible to kind of pin these LLMs down into what words are acceptable and which aren't.

Jamie Bartlett (17:15)

Yeah, exactly. It's not really about words, is it? It's your. The jailbreakers are manipulating them. Psy. Almost moving them around and sort of making them forget things, confusing them slightly. And knowing how you fight against that as. As the techniques get more sophisticated is quite difficult. You couldn't ban the word bomb because there are so many legitimate uses for it, but also the companies wouldn't want you to because then the model becomes very, very unhelpful and then not profitable for them. And it's evolving. So when someone like, let's say I filed a jailbreaking report to ChatGPT and said I have managed to get it to make this racist essay, they should say, oh, I see, what this person did was manipulate it by using a play and then gradually walked it through. I wonder if I can retrain. We can retrain the model slightly or tweak some parameters so that the model's more likely to spot that. That's how they do that. They don't ban words. They say they look at the techniques and think, oh, I could maybe we could recognize the motives of the user better. So we'll block it that way. And it's important to remember that if you jailbreak a model to give you a racist essay or something like that, a jailbroker model that does that won't then just give you biological weapons. Each one is separate and different. And to get to the hardest stuff like biological and chemical weapons is much different. I couldn't do it in a million years. It takes people like Valen days or weeks of like non stop even days or weeks.

Annie Kelly (18:55)

Doesn't sound like that long.

Jamie Bartlett (18:57)

No, it doesn't sound like that long, does it? And people are worried. And there's companies like Far AI that specialize in this and they file reports regularly to the companies. They do say there needs to be more transparency from those companies to show like their progress to say what they're worried about. And there's not really any formal reporting system like if you as an independent researcher. Some independent researchers have tried to file jailbreaking reports to these companies and they've just kind of been ignored.

Annie Kelly (19:24)

Wow.

Jamie Bartlett (19:24)

So there needs to be some more formal ways that people can do this because at the moment, jailbreaking is one of the best ways to try to keep them a bit safer. It's not that it's not perfect and unfortunately it is a double edged sword because in some of the forums where this is discussed, obviously there are people there thinking, oh, I could use this to automate a hack. Oh, I could use this to make some more propaganda. So it's, it is really difficult.

Annie Kelly (19:53)

Coming up, are we heading into a future of jailbroken AI robots?

Kai Wright (20:13)

I'm Kai Wright.

Cari Sherman (20:14)

I'm Cari Sherman and we are here

Kai Wright (20:16)

to tell you about our new show, which is rooted in this feeling that at least I have. I know you have, where, you know, it's kind of like when you wake up in the morning and you pick up your phone and you're just hit in the face with a fire hose of news. Right?

Cari Sherman (20:32)

Like there's war, there's authoritarianism, our planet is burning. I could go on and on and

Kai Wright (20:37)

on and on and on and on. But like, we're trying to figure out how to manage it, right? Like, how do you manage it?

Cari Sherman (20:42)

I manage it by leaning in and trying to learn more and trying to figure out, okay, how can I be smarter about this particular topic and who can I talk to that's going to make me feel better about it and

Kai Wright (20:53)

who can tell me who's responsible for the mess that I'm reading about? So that's our mission. That's the show.

Cari Sherman (20:59)

Welcome to Stateside with Kai and Carter. We're a new show from the Guardian.

Kai Wright (21:03)

We're talking to big thinkers and the best journalists just trying to understand the world through smart conversation and honest reports.

Cari Sherman (21:10)

We don't have billionaires telling us what to say.

Kai Wright (21:12)

Stateside with Kyan Carter will come out three times a week, Monday, Wednesday and Friday, starting May 13.

Cari Sherman (21:17)

Follow on Apple Podcasts or catch us wherever you watch or listen.

Annie Kelly (21:24)

So far we've talked about these jailbreakers who are seem incredibly sophisticated at being able to want to test the limits of those safety features and push past them. You know, people like Vailen seem like they're doing it for good reasons, you know, automate money or whatever. But there is a darker side to this, isn't there? People who want to Crack open a model for more criminal or nefarious reasons.

Jamie Bartlett (21:46)

Yeah, of course there is. I mean, imagine how powerful a jail break to automate or to design brand new malicious software, malicious code, ransomware. This is very, very useful for a criminal, obviously, and there are rules and safety filters to try and stop them. And criminal groups work out ways to get around those. Because if you can automate all of your phishing emails and you can automate all of your stolen data dumps being processed, that's very useful for you. On the darknet, people claim, I haven't tested it, but they claim to be selling jailbroken models either. Like, we'll give you access to a jailbroken model we've built because a lot of the jailbreaks are on open source models which you can then share with other people. So give you access to our jailbroken open source model. Or maybe here's a series of clever prompts that if you use these, pay for them, use these, you will be able to automate your phishing emails or get it to write loads of phishing emails for you.

Annie Kelly (22:45)

So you've broken this, you've got this jailbroken bot that you're then able to sell on and monetize. Because as long as you know what prompts to use, it's going to do,

Jamie Bartlett (22:56)

or you're selling access to it, you're selling access to one that you have built. Like I say, people fine tune open source models that are available and say, right, I've kind of managed to dismantle a lot of its safety filters. Go for your life. Now they get patched up regularly so they don't work forever. So it sort of all depends on when the companies update them or figure something out. And it is it. I mean, talk about cat and mouse game. And I know it's always been that way in cyber security, but it really is like patching them up, finding out a problem, trying to patch them up again. And it's just, I mean, it's never going to stop. No.

Annie Kelly (23:31)

What's even more frightening is that this is where we are now is that, you know, AI is already very far away from just being a chatbot. We see more and more of these AI agents, you know, they have agency, they're able to get access to a lot of personal data and do things like write emails for you or make bank transfers for you, not just open information out there on the Internet. How could that impact the future, do you think, of jailbreaking where we are

Jamie Bartlett (23:56)

at the moment with these models, they generally still are Chatbots. For most of us, they're producing words, but they can contained almost within that virtual world. But increasingly there's this mad dash to turn everything into an agent. People create bots on top of their language models which are given access to their, I mean, their bank account sometimes or like, or at least like the crypto wallets or their emails or the calendars and are given tasks to do, like, you know, I want you to make me more money, I want you to try to find, you know, you send all my emails and automate, automate, automate. And it's getting more and more, more intense. Now. If you start jailbreaking models which are agents that are out in the real world doing physical things, running software on robots or whatever it is. I know it sounds sci fi, but a jailbroken physical robot running off a large language model that would then do things that you told it would be. Can you imagine what?

Annie Kelly (25:00)

No, I mean, it sounds like a Terminator or something.

Jamie Bartlett (25:02)

Doesn't sounds a lot like the Terminator. Yeah. So the reason some of this safety research is so important and one of the reasons I wanted to get it out now is because we're entering into a world where these language models have real power, like physical attributes in the real world that they're doing physical things. And the more powerful they get, the more dangerous a jailbroken model would be. I think they're getting harder to jailbreak, but they're getting more powerful. So when they are jailbroken, they're more dangerous.

Annie Kelly (25:35)

But what do you think, or what did the jailbreakers tell you needs to happen to make these systems safer for all of us?

Jamie Bartlett (25:44)

I think all of them tend to agree that the companies aren't really investing quite enough money or effort into it. Particularly they don't test them enough before release. I mean, I think you shouldn't really be able to release any language model into the world unless it's gone through some kind of independent, rigorous testing, not run by the companies, but run by a, some kind of government agency that runs all these tests and says, we're broadly happy they're quite safe, but I am not that optimistic that it will happen until some very bad thing happens first and forces everyone to act.

Annie Kelly (26:20)

Jamie, thanks for coming and sharing this dystopian of the future. Thank you very much for coming in.

Jamie Bartlett (26:26)

Thank you.

Annie Kelly (26:30)

And that's it for today. My thanks to Jamie Bartlett and his book how to Talk to AI is out now. This episode was produced by Guy Zafman and presented by me, Annie Kelly. Sound design was by Brian McNamara. And the executive producers were Homa Khalili and Sammy Kent. And before we go, I just wanted to tell you about a new video podcast that our New York office is launching. It's called Stateside with Kai and Carter. And it's hosted by our colleagues Kai Wright and Carter Sherman. And each week, they're going to be trying to make sense of some of the biggest stories happening right now. The show will feature conversations with some of the smartest thinkers and reporters, not just from the Guardian, but across the world. It's launching on the 13th of May, with episodes every Monday, Wednesday and Friday. You can find it in full video on YouTube and wherever you get your podcasts. And we'll be back later in your feeds this afternoon with the latest. This is the Guardian.