EP 027 The Dark Trade In AI Child Sexual Abuse Material: A Discussion With Journalist Octavia Sheepshanks

New article Nov 2023
https://thecritic.co.uk/the-shadow-side-of-open-source-ai/
Octavia Sheepshanks is an intrepid freelance journalist and researcher driven to uncover compelling and demanding stories. Her diverse subjects of interest include space travel, mental health, feminism, and AI. Octavia’s current focus is on the dangerous proliferation of AI-generated child sexual abuse material, and she has dedicated herself to extensive research on the problem throughout 2023.Her work first came to prominence through an investigation conducted with The Times, which provided a sobering insight into a subject that demands immediate attention.In a subsequent investigation with the BBC, in which Octavia exposed the growing trade of illegal AI-CSAM on Patreon, and highlighted the dangerous consequences of Stable Diffusion’s open source release. In continuing to shed light on the creation and trade of AI-generated child sexual abuse material, and the terrifying scale of the problem, Octavia hopes to contribute to global efforts to fight the issue.
The Times https://archive.is/P9t8B
BBC News https://rb.gy/qjeno
BBC TV https://rb.gy/wf2xk
Mostaque on ethics https://rb.gy/u90c5
Bakz T. Future Statement https://rb.gy/4k023
5Rights https://rb.gy/j8nn4
NSPCC on E2EE

Security Circle ⭕️  is an IFPOD production for IFPO the International Foundation of Protection Officers

Transcript

0:00

Hi. At times. Here at the security circle. We've had to advise that some content that we'll be talking about. Can often offend or otherwise cause discomfort to some listeners. The following program that we have put together today. Does contain material that may be disturbing. It does contain content, which may be too intense for some viewers. The following program does contain. Sexual content and themes around sexual abuse. So listener discretion is advised. When I was in the police. In a. A very junior role. It was my job once to attend a very well known branded. Beverage company. To investigate a report of a number of indecent images on an employee's computer at work. And these sorts of jobs, they didn't come up very often at all. This particular business had hired a private investigator. The private investigator had in fact, confirmed that criminal offenses were disclosed. And this is how the police came involved. We arrested that employee took him to custody. He had in his work locker. A pile high up to certainly above my waist. Of a number of different printed documents involving stories of incest. And it transpired on his work PC. He had in excess of 3000 images. That we're a very indecent in a range of different categories and they all overall very young children. He was processed in the normal way. He had a lawyer. And the lawyer came in and supported him through the process. I had to go to his family home. And when his wife answered the door, I was fully aware that I was going to be ruining her life. She was obviously concerned that somebody at uniform was at the door. I had to sit her down and explain that I needed to. Take away his computer laptop. Mobile phone, everything that could potentially have images and search the home for any other content that he might have hidden. And I had to explain to her why he was arrested. And from that moment, I was fully aware that I had changed her life. I had changed her world because she would have felt incredibly isolated. She wouldn't have been able to tell her friends. She would have probably felt a lot of shame brought onto the family. And I just wanted to give that perspective. On how, what we're going to discuss today. Does destroy people's lives and it destroys people's families and it destroys people's homes just over a week ago, the BBC launched a story. A story that involved AI generated child sexual abuse material. And the story was broken by a freelance journalist called Octavia sheepshanks. The story was so compelling and we all know about the risks that AI presents, and we all know about the amazing ways AI can support, especially at work. So I got in touch with Octavia sheep shanks. Knowing how much the BBC would have had to have censored her story. And I asked her if she would tell me her story. And on Friday, we sat down and we chatted for two hours. And what I'm bringing you today is her full. Account of the last four months. Investigating what is certainly we would all agree. One of the darkest areas of our human existence.

Yoyo 4:10

Hi, this is Yolanda. Welcome. Welcome to the Security Circle podcast. If PO is the International Foundation for Protection Officers. We are dedicated to providing meaningful education and certification for all levels of security personnel and make a positive difference to our members mental health and wellbeing. Don't forget to give us a like on whatever platform you are listening to us on. It really matters to us. Also, if you check out our website, www.ifpo.org, you can sign up and receive our podcast each week and never miss them. Our audience is global, so wherever you are around the world. Thank you so much for joining me.

4:52

My very, very special guest with me today on the security circle. Is octavia sheepshanks Octavia's current focus is on the dangerous proliferation of AI generated child sexual abuse material. And she has dedicated herself to extreme research on the problem throughout 2023. Her work first came to prominence through an investigation conducted with the times which provided a sobering insight to the subject that demands immediate attention. In a subsequent investigation with the BBC in which Octavia exposed the growing trade of illegal AI child sexual abuse material on Patrion. And highlighted the dangerous consequences of stable diffusions open source release. In continuing to shed the light on the creation of trade of AI generated child sexual abuse material. And the terrifying scale of the problem. Octavia hopes to contribute to global efforts to fight the issue. And this is her story.

Octavia 5:57

Octavia, thank you so much for joining us today. Look, AI cap, G p T. There have been an awful lot of headlines around ai, some of them, most of them, quite fear mongering. What we're gonna talk about today isn't very comfortable listening at all, but it's a story that's worth telling. First of all, just frame it, how did you get involved in writing this article about how AI is using imagery to represent children in, I don't know, Octavia, how do you say it? Yeah. Well, I've just been calling it Ai C Sam it's difficult because child pornography is a phrase that the organizations now are trying to dissuade people from saying, aren't they? Exactly. Of course. There are technically, if there are no children being harmed, but personally, I still think, well, it's an image of a child being sexually abused, so it's still not appropriate to call it child pornography. But then, it just ends up being so many words. If it's like artificially generated child sexual abuse material, it's like six words long. But anyway, so I'll just call it AI csam, which sounds right. Yeah. How I started researching this was because last year, so July last year, I heard that mid journey, the AI. Image generator, was releasing or opening up into beta mode so anyone could join and start testing it and kind of contribute to its development and growth. And I was very curious and, immediately signed up. And so the day it became available, I was on there like coming up with, random ideas or, typing in prompts and things. And for me, that ended up being, making images of baby dragons. And so periodically I would go back on to mid journey, which is the most. Kind of user-friendly, uh, AI image generator probably. And also, uh, very it wasn't originally intended to be realistic. It was meant to be kind of artistic. So, yeah, but that, I mean, it's just improved the fact that I've been using it kind of since then and being able to watch it get better and better. I mean, it's just been absolutely mind blowing because it can just create, whatever your, almost like your heart's desire, what you want to see. I created a character of his baby dragon living in London, and he moved to London from his dragon world and he's quite like boring and eccentric and. Yeah, I kind of made this little Instagram account of him and I don't think many people kind of got it or found it as funny as I did. But yeah, it was just something interesting to play with cuz I'm seeing myself as quite a creative person, but I'm not an artist. So, to have something where you can have these random ideas and then play with them and it almost is like an imagination kind of mentor. So I personally, I think all the concerns about, artists and the threat to artists, I don't really feel that way. I feel like artists should definitely have a go with it themselves because then, you either discover like, oh, this can't do what I'm doing anyway, so it's not a threat to me. Or actually this will save me some time doing some aspect of my work. Oh, what am I gonna do with this 24 hours that I didn't have before? Which I think is how coders see. Chat G P T, they're like, oh great, this can do all the boring work. And now I've got even more time. But yeah, I mean I completely understand why artists feel the way they do, but to see AI image generators as AI art is quite odds. It would be like calling chat G B T AI literature or something, or AI novels as if that's the only use for words in our society. So yeah, I was just kind of been messing around with it for the past year, like on and off, and I actually had a literary agent interested in potentially making a children's book out of my baby dragon that I thought actually, yeah, I thought, okay, I'll just get a lot of hate for this. And the kind of copyright wasn't. It was still all very vague and constantly shifting and I, but actually at the time I was very interested in children using mid journey to kind of help manifest what's in their own imaginations, because of course, children's imaginations are so much greater, they're greater than ours, and yet their actual skill in art is just inevitably like, stick figures and whatever. So they can just imagine these incredible fantasy worlds and can't bring them to life. So, yeah, I had a go actually with Some children on the bus who saw me messing around with a baby dragon and, were so intrigued and their parents were like, okay, fine. You can, have a go. And yeah, I just thought, wow, like this is gonna be a really big thing. But I just had no idea what other people were doing with it. So then in April, so with Mid Journey and yeah the B B C story was all about stable diffusion, but the journey of me kind of realizing what was going on, started with Mid Journey and I had a piece in the Times about Mid Journey. Mid Journey, it's not open source, which means that the company, which is a really small lab they. To have control over the product and the platform and how it's used. And you can only use it through their website, which is, well, it's a sort of add-on through the platform Discord. So you have to have a Discord account and you actually have to pay to use Mid Journey as well. Partly because, I mean, I think the, actually this is a whole separate thing, but the energy use is just like colossal. Anyway, so with that, they actually have a public gallery for anyone with an account, so you can see what images other people are creating. And from the beginning, I hadn't really spent much time on the public gallery because I felt quite depressed every time I clicked on it because it was just, images of young, attractive women just almost always, or kind of game characters Because there's so many images being made. I mean like thousands every second or whatever. I mean, there are, I think, yeah, I'll check something like the 16 million users of Mid Journey or something like that. Wow. Basically with Mid Journey, the company really tried to restrict. The use of it for anything pornographic, even anything violent. And the way they were doing that was through restricting the words people could use. So you can't say breasts, you can't say sex. there's so many words you can't use. And the more people were trying to use certain words or get around it, the more words they would block and the users were being, very angry. And yes, there are more than 15 million users of the Discord server. Some of those may be duplicate accounts, but yeah. The reason, the way it's become so good so quickly is because so many people are training it, and most of those people are men. And I think it's very worrying that the tools of the future are essentially being trained according to the preferences of. Whilst very specifically young men who are mostly from the Western world or East Asian I think the platform became much bigger in China. and then that started to shift the kind of images of women that were being made and there were more East Asian women being generated. But, it's not I think what Mid Journey should have done and what other image gen AI, image generator companies should have done is to select globally users, all genders, races, ages socio-economic backgrounds and used and, let them train the models because. The model now has a specific understanding of the word woman, and it basically means mainly white, young, very, like conventionally attractive. that's a huge problem. But I actually, at one point earlier this year was messing around trying to make, I did a kind of sci-fi series of women around the world or in different kind of natural locations holding, glowing circular objects. And I kind of imagined that it was a similar world to ours, but that women had special ways of communicating with each other through these different objects. And so I was, yeah, just Making those, and, trying to get the women depicted to be more ethnically diverse in different ages and all of that. And so I was kind of battling against the algorithm, like desperately trying to, Make the, these characters I was creating more diverse. And yeah I find, yeah, the whole thing very interesting. And I will yeah, show you some of those pictures and they're completely photorealistic. Like, you literally cannot tell that they're not real people. I mean, the, there was a time when newspapers were reporting, oh, don't worry, everyone, AI can't do hands. And the window in which AI could generate photorealistic images, but with weird hands was literally about six weeks long. And then Mid Journey released. Version five, and then, oh, look, it can now do hands.

Yoyo 16:12

So this technology, this AI technology is learning exponentially, and it's taking me back actually to your point around, one gender almost designing how this technology learns, but not just one gender, one specific stereotype of gender and the, there are dangers linked to that, which we'll go on to. It reminds me of a gender inclusivity in technology qualification that I got, recently in the last couple of years. And we learned how. For example, do you remember when they got the algorithm wrong with recruiting, and they were only using men's cvs to attract keywords. And so any woman's CV was not being picked up. And that was a typical example of a huge learning because only men were involved in creating that process, creating that algorithm, and creating the basis for which they set the precedent, which is why it's important to have a diversity in the creation of any technology. So that one or the other isn't biased negatively. So actually it's quite scary that you have a very specific. Group of people. Yeah. Demographic with a not great intent either. Literally designing how some technology is going to be developing. That's phenomenal.

Octavia 17:38

Yeah. Yeah. So that was a huge concern of mine. And also the primary interest of that group of men is gaming. And so that's just a specific world. And if you look at the stats of the visitors to the Mid Journey website and users of Discord and mid Journey is by far the largest server on Discord. So on this site, SimilarWeb, you can analyze kind of website traffic and the primary other interests of mid journey users. It's said gaming and porn, and you can see that it's, mostly male, maybe 80% and if not more, and maybe something like 16 age range, 16 to 35. So anyway I thought, okay, that's a big problem. And normally I write opinion pieces, so that was a story that I wanted to write and I was kind of pitching that story around and it didn't get picked up. The headline, would've been, it's a problem that the image tools of the future are being trained by this specific demographic of young white men. So then I thought, okay, I should, I have a look more, like, more deeply at what images are being created. And this was all through mid journey because you can see, directly what's being made. And I knew that stable with effusion, existed. And, there were other image generators too, I was only researching it through mid Jenny. And with that you can't make any images of Nity. I mean, people will try, but it won't generate nudity. Having said that, it had basically been trained to generate images of naked women's breasts purely through the training. So even without using that word, whenever kind of breasts had popped up by mistake, the users had chosen to upscale that particular image out of the grid of four that gets generated when you type in your prompt. So, if that's basically just how the model gets trained. It thinks, oh, okay, woman breasts great. And it will do that without you even needing to use the word. So whenever mid Jenny would release a new model or an updated version, it would be even, it would lean even more heavily towards, youth naked breasts, et cetera. The other thing to add is you can also use images as prompts themselves. So the AI model is trained on nearly 6 billion image text pairs. So there's this huge data set called Lyon five B, which was released open source, and that is essentially a huge kind of coded, sort of blob of the connections between images and words from the internet. So it's, no, it doesn't work through having, the machine doesn't have all those 6 billion images and create some mashup of it. They've actually been trained to understand concepts of their own, of its own accord through the pairings of words and images. So, yeah That's essentially how it works and that's why there've been loads of different image generator models created kind of simultaneously because of the release of this data set, Lyon five B. Actually the people that release that set are also, kind of to blame in a way for everything that we'll be talking about. I then thought, okay, I need to see more specifically the images that being made. So I went onto the Mid Journey gallery and instead of clicking off it as I'd done in the past, just started going through, all the images that have been made in the past few minutes. And yeah, it was just all. Know women, blah, blah, blah. And then I started to see more of essentially women with white liquid all over their faces. And the prompts that had been put in would be, young women with yoga all over her face or, because of course you can't say Yeah. Or whatever. You have to say that or you can't say semen yeah. They were kind of just trying to get around it essentially. And then I saw that a lot of users were uploading images of real women using those images as prompts. Oh yeah. so what I was gonna explain before is that, you can use images as prompts in themselves. So you've got the model which responds to text prompts, and then it, you can put in, say on mid gen, you can put in up to five images and create a kind of, Mix of those, but it can do that in a very creative, interesting way. And that's something that I find really fascinating anyway. So what these men are doing is uploading images of real women and then trying to generate them with semen all over their faces. And you can then go to those users' profiles and see that they are generating this on an industrial scale. One user had his whole profile was dedicated to this one woman in loads of different places, oh, public toilet with milk all over her face. And this was just all out there on the open internet and also Mid Journey claims to be a friendly, child friendly platform for over thirteens. And I then found that, Loads of images of children were being generated and images of real children used as prompts. And that really concerned me, obviously. And people using prompts, kind of getting around the words getting around the prompt bands, by saying things like, young baby with father in sauna or something. And then the, you could always tell that what they wanted to generate was kind of restricted by the word prompt, but the images that came back were still, very disturbing. Using words like non consensually, young teen pushed to ground, or, all these just, yeah. Very horrible prompts. And so that, Led me to work on an investigation for the Times, specifically about Mid Journey and Discord, who were facilitating this and had previously, so Discord had previously deleted servers, which were dedicated to creating deep fake porn, and they had said they had a zero zero tolerance approach to this. But there are millions of servers on Discord and mid, and the average server I think has about 20 people in. But Mid Journey has a, I've said, over 15 million users and is by far the biggest server within Discord. It's almost bigger than Discord itself. And so I thought, well, if Discord were to take the same approach and be consistent, they should technically be deleting their. Biggest server, which is actually a huge company in its own right. But they just said in response to the Times piece, oh we take a zero tolerance approach, this is bad. And the piece was on page 12 and obviously the Times is behind a paywall and yeah, they just got away with it and mid Journey just carried on. And yeah, so that was the first story. Might be worth asking you then around certain images being used as prompts. Is that because those real images, do you think they were illicitly obtained in the beginning? Is there something about them that's particularly, relevant to them being used as prompts and in relation to what they generate? Yeah, so the images that I could see that had been used as prompts of real women and girls, it looked like they had just been, got off the internet or maybe off dating apps, or, I mean, one of them, the screenshot, it was clearly a screenshot of an iPhone, uh, photo gallery of a girl smiling happily on holiday and then. The user desperately trying to generate her in sexual poses and you just think it's probably not like a consensual thing. And then the one that I mentioned, which, this whole profile was dedicated to generating images of this one woman in kind of compromising positions in public with, milk all over her face. the picture was of just, it was just a smiling selfie in a mirror, in a hallway. And, you dunno where those images come from, but essentially just yeah, on the internet and then the images used of children, they would often be clearly kind of just amateur photos of people's own children, like children of different ages, would all have been uploaded and used as a kind of mashup, uh, and. I mean the volume of images of children being created was just huge. And they were all, they'd come back mostly clothed, but, users were spending days and days generating tens of thousands of images, each of just random toddlers. And yeah, that was obviously very concerning and it made me think, if I have children, I'm not putting a single image of them on the internet ever. Yeah. And that, made me think about people that I know who put images of their kids Yeah. On Instagram and everything. Yeah. And, the images I that I was seeing that were being uploaded were very clearly just, random children. Not just like child models and stuff. And I remember, saying to, The tech correspondent Marksman, who was amazing and worked on the Times piece with me because I hadn't worked on any kind of investigations before. So I'd contacted him directly and said, I normally write opinion pieces. Would you work on this story with me? And he was great and led the way, but let me do the well, I'd already done most of the research anyway. So he essentially put together what, I think I had a 60 slide long research document and he, put it into this. News piece, which was great, but I remember saying to him, oh yeah. But you know, this is all about mid journey. But I know that stay with Fusion will be so much worse because yeah, it's open source and you can use it offline and there's no restrictions on words you can use. It feels really weird going after Mid Journey because they at least are trying to reign it in even though they're failing. They're actually trying and he just said, look, we've just got to work on what we have the evidence for right now. And the whole point about stable with effusion is you don't know what's being made because you can't see it. So we can't really write about that right now. And that was helpful cause I just thought, okay, we'll pursue this for now. I actually didn't realize that I would be able to research stable diffusion because I didn't expect all these images to just be floating around on the Surface web, which it turns out they are. So the

Yoyo 29:50

drastic differential here is that stable diffusion is open source. Okay. So let's talk about the difference. So you've got mid journey, you have to subscribe, you've gotta put your details in. Yeah. Yeah. So the difference between closed source and open source, or the narrative at least, is that, ethical businesses or ethical tech companies, Release stuff, open source, because it means that they're transparent about the, what's going on. It's not a black box. It means that smaller businesses can then use that technology and build on it themselves. It's, a kind of it's seen as, it's like releasing the patent for a vaccine or something like that, yeah. Intellectual property. It's also seen as very Kind of, it's seen as very financially selfish to keep your product closed. Sourced. Yeah. Cause it's seen as, fo being focused on money rather than releasing technology to everyone. It's seen as contributing to a world in which huge companies have monopolies and, that's it. So there's this sort of open source philosophy basically that's the right thing to do. Yeah. And so companies that release anything open source will get a huge amount of kind of street cred appreciation companies, which don't get, a lot of hate and and so OpenAI, which made chat to pt, they closed source. And, they get a lot of criticism for that. I think at one point they said they were gonna release it, open source, then they went back on that, which I am very glad about. It's interesting that's the narrative because personally I think with ai, when you release something open source, you are essentially, opening Pandora's box and there's absolutely no way back. And you've then put something out into the world that can be used for terrible criminal activities that can really harm people. And that's happened. And so that's what's happened with the stable diffusion. There's a whole story there that I find absolutely fascinating of this one man last summer desperately trying to at least delay the release of stable diffusion, stable Diffusions source code. And he. Essentially failed. But he released all the chats with him trying to delay it, pleading with the c e o to delay it, talking about all his concerns. He got so much abuse from within the tech community that he had to go off the internet for a time. And actually the, the code for stable diffusion was actually leaked on four chan a few weeks before it was made public anyway, which is also its own risk. So with chat G B T, it could technically be it could be leaked as well., it's not just up to them. Someone could leak it out onto four channel or Reddit or whatever, and then that's it forever. So we can't ignore. Certainly I can't, with my experience as a detective, that there can be some very sinister motivations into this being blocked because people will understand and know what its potential is. it's not just, is it the ethics around whether IP should be shared and whether it's ethical, whether there's efficacy in sharing everything with the general public. These people aren't stupid. They know that this technology can be used for a Mull intent. Yeah. So they were actually huge kind of ethical and philosophical discussions that went on before stable effusion was released publicly. And they're just fascinating to read. So actually maybe I could read a few of them out. Yeah. We should probably provide some of that source material on your bio as well, so people can click onto it. I would love to read out the part of this blog post from, yeah, do it. The man who tried to delay the open source code. So his online name is Baxt Future, and he is a software developer who's been monitoring the development of AI for years and yet received heavy criticism and abuse from the AI community in 2022. After trying to persuade the c e O of stability ai, which part developed and then. Owns slash released stable effusion. The C e o, Amad Moak, his name is, which I actually think sounds kind of like I made a mistake I realized today. Anyways, so Amad Moak, Emma Boak is the c e o of stability ai, the company that develops the stable diffusion model and released it publicly. So BT Future has a blog post called Statement on stable Effusion. And it goes through the whole story of him, desperately looking ahead and trying to prevent this thing from being released. And he even also put out a. Google Slides document with all the screenshots of his messages with IAD Stack. And in fact, maybe I'll just read out this whole, yeah, do few paragraphs that I've got. Yeah, Reason for this is that stable diffusion is gonna become the very key part of our conversation. Yes. And what we're gonna talk about is really horrible. And I have to put a disclaimer at the beginning of this podcast because it's that disturbing. So the fact that there were significant moves to try and prevent what we are about to tell you is why it's really important we focus on this part right now. Yeah, exactly. And it's also really important, I think, To expose the dangers of open source, especially when it comes to ai. Yeah. And personally, I think the benefits of the open source philosophy and the open source movement don't apply when it comes to something, this powerful. And yeah, I think even though it's too late for image generation, it's not too late for other forms of ai. And, if chat G B T, for example, were released open source, that could just lead to huge criminal activity because it would make the average person so much more intelligent for example, you could have. Children learning how to make nuclear bombs or any, anything that, that kind of thing. And I'm just, things like normally on Google you can't google how to dispose of a body. But with, if chat G B T were. Off. If it were open source, you could have it offline. You can train it on whatever you want. You could train it on every criminal, what every criminal has ever successful criminal has ever done. Yeah. And then you could get a tailor made answer for how you should dispose of your particular body of whoever you've heard it, and not get and get away with it. So yeah, just all sorts of things like that. So I, although you remember in the past, Siri used to be able to do that. Oh, really? Used to be able to ask Siri how to get rid of a body. And it was all tongue in cheek, obviously with the makers of Apple. But they would say, Siri would come back and say, here's a list of reservoirs, swamps, forests, woods. Oh, really? Which is why I remember this list comprehensively, because we used to ask it quite a lot. So, uh, where can my buried dead body, Siri, here's a list, here's a list of, barren land. It was all a joke but yeah, but this, yeah, it wouldn't be a joke because you could put it in, the temperature, the location, the, yep. Yeah. Anything you wanted and, oh, scary. It would be cleverer than the law enforcement trying to look after this. Because the thing is that the, once something's open source, you then train your own version of it. So the police in that kind of scenario wouldn't have access to the tool unless they had trained it themselves. So, yeah, there's also things and yeah, I just don't really understand why when people are talking about their kind of AI concerns and AI blowing up the world, they're not actually focusing on, what humans could do if they had unmitigated access to ai, which they do when it comes to image generators through stable effusion. So, This is an extract from Baxt Futures blog post statement on stable diffusion. And this is after he had, it's after stable diffusion had already been released open source, and he's explaining why he spent so much time trying to delay it. And, okay, after this blog post, he then says he needs to take some time off the internet to look after his mental health cause of all the abuse he got. So he says throughout the various stages of stable effusion. Oh, yeah. So to clarify this was all operating within the stable diffusion discord server. So it had been trained privately. Uh, it wasn't, I don't think it was open to everyone, or at least it wasn't publicized. So there were, I don't know, probably tens of thousands of people training it on. The stable diffusion discord server. Throughout the various stages of stable diffusion, I've received dms and calls from people who were concerned by the offensive and horrific content they were seeing on the stable diffusion official discord server. Photos of violently beaten Asian women were the kinds of P images people were originally generating on stable diffusion at a frequent scale. This is even knowing those results would be public for all to see and associated with their discord usernames permanently. Many of those people have told me they are still disturbed by what they saw. Eventually, I believe with the help from moderators, stable effusion had cleaned up most of the disturbing and or unethical images they were seeing. However, I was now getting additional messages about the inherent biases in the model. Someone sent me a screenshot where they had simply entered the monkey emoji into stable diffusion and got back photos of black children. Was there any work at all put into bias before releasing this model to the public? I have to admit that I found it peculiar in the few weeks of its existence. The stable effusion community was rampant with offensive, disturbing, and or horrific content, and evolving quickly. It was simply too toxic. And without the help from their volunteer moderators, they perhaps wouldn't have never recovered. Ahmad eventually actively discouraged, not safe for work content inside their discord server. And I believe the community had made written strides against generating racist and offensive content. And this is a key quote for me, Octavia. However, I just don't understand if it was too toxic, illegal, and or dangerous to have on their own turf without heavy active human moderation. Why were they releasing this model right away onto the whole world's front yard? By my understanding, the model is capable of generating illegal child abuse images as well. Is this also encouraged by their organization under the not safer work label? At times, I see that Ahmad is discouraging this publicly on the Discord and the licensing terms do not allow for illegal content. However, by releasing the model, his actions say a lot too. So I just don't know what to think. So he's said the cat's out the bag now then. Really? Yeah. He's, he wa he warned of its dangers. He warned the lack of governance and now his worst fears have come true. Yep. And this blog post of which that was a quote, includes a link to a 36 page Google Slides doc containing screenshots of his exchanges with Amad Moak, in which he expressed ethical concerns, which were minimized and mocked by Moak. It begs the question, and it's easy for us to say this as women, and then people would stereotype us for being opinionated because of our gender, but why is the community so eminently defending something that's just become so goly? I don't expect you to have the answer. No, I think I do. I mean, I think because they see it as a tool that is so universal, so, As I said before, it can create whatever is in your mind's eye, you can bring to life anything that's vaguely hovering around in your imagination. So the people that were defending it had already been using it for their own kind of desires, so their own creative projects or gaming projects or whatever. And I have to say myself, if someone said to me, oh, image generators are not, are just not gonna exist anymore, I would be really upset. And I don't mean, open source state, but diffusion, I mean something like mid journey. I mean, yeah, just because there are just so many things that I want to do with it that I haven't even kind of got around to doing. So for example, one of my favorite children's books. Are the Cresta Maney series by Diana Wynn Jones. And there's never been any TV adaptation of it. So what I really want to do is generate film stills of exactly how I imagined those characters to look. And you know exactly how when I was growing up reading those books obsessively and I still read them how I see the characters. So yeah, that's just something I want to do. And then other things like designing prints myself and then getting them made and then making my own clothes with prints that I've designed myself. And those are things which, an individual does not commission artists to kind of do their bidding basically. Screenshots reveal. This is, yeah. Something I've written. Screenshots reveal that when Bax raised his concern of dealing with the aftermath of the model for years to come, Mustak responded with yes, but the net benefit will be 99% positive. When asked how he could make such a calculation, what Slack said, it's because I believe people are inherently good. Whereas AI ethicists believe people are inherently bad. I also say this as a qualified religious scholar who helped set up the university for cross-cultural ethics at the Vatican. Bacs posed a specific scenario. So if a celebrity gets depicted negatively and in a not safer work context, stay with a fusion image and kills themselves over body image issues, does your math still check out? It's worth it. To release a model prematurely must act responded with, well, how utilitarian do you want to get? So, yeah, I'm a whole Stu. Yeah. Uh, B then says, finally I want to say this is the last I have to say on this issue. The weights already out there and the world has chosen to go in this direction. There is nothing more I can do at this time. To be clear, I've found all of this not only exhausting, but in fact nauseating. Except for major AI announcements. I'll be taking a break for a few weeks from YouTube, Twitter, and other social media. I hope to come back with a renewed perspective and some new content that is more positive and just gets back to sharing the kind of stuff I enjoy. I started my channel to share my interesting ideas about the world to the audience. I'm sorry for the negativity coming from my Twitter and other channels lately, and the absence of content on my end while dealing with all of this. So I've written in my notes about this. Clearly the narrative around open source and the tech world and even more broadly is so positive that questioning any element of it can lead to a torrent of abuse and criticism. I'm just wondering why he used the word utilitarian. Why do you think he used that word? He says utilitarian a lot. I think. I dunno. I think that's just his approach. That's his ideology. He seems very kind of blase about the whole thing. There's another interview with him, released soon after the open source release where he says something like, people ask me. What do you think will be made with this? And I'm like, I don't know. It's a primitive, oh, he also speaks about how he sees it as a problematic, ethical attitude to blame the tool or blame the creator of a tool for what it then creates. And he likens it to the ethics behind some laws in Islamic countries, which ban women from driving cause of what they might do. So the logic that I'm not a fan. I'm not a fan of this guy. I wasn't, anyway, no. So less of a fan now. Yeah, I would absolutely love to have a discussion with him myself. So if you want to invite him on the podcast for a discussion with me. Yeah. which you might say yes to, because he's always very responsive to, criticism and will engage with what people say it about him. So he says, although some images made with his creation, maybe unsavory, so he told Wired on the subject, using technology has always been about people's personal responsibility. If they use Photoshop for illegal or unethical use, it is the fault of the person. The model can create bad things, only if the user deliberately makes it do so. So this is the principle of the nra. In relation to the Second Amendment, they're basically saying it's not the gun that kills, it's the human. Right. Exactly. It's just the human can really struggle to kill someone if they haven't got a very powerful machine gun that sends out several thousand bullets every few seconds. Yeah. Okay. I think our conservatism is fairly aligned there. Alright, let's talk about stable diffusion then. Now we understand the precedent. Who's using it and what are they making? Octavia, I, so I think it's mainly used by men, young men whose primary other interests are gaming and porn. I suppose Porn Hub is the 11th most popular website in the world. I believe. So, I suppose it's a popular interest but I think. If the kind of a discussion of the dark trade in AI child sex abuse imagery with journalist octavia sheepshanks imagery that people want to generate with AI is the kind of imagery they're consuming anyway,. So what are the dangers of artificial child sexual abuse material on this particular platform? Because you can talk to that through experience..

Octavia 49:20

Well, I'll say first that yes, stable diffusion isn't really a, it's not a platform. It's a technology that you can download. It's code that was made freely available by the company, stability, ai. So now anyone can download and install it and run it on their own computer. in theory it only takes up around four gigabytes of space, but in practice you'll need more. Ram basically like working memory, for it to function effectively.

Yoyo 49:52

So, because stable diffusion is open source, that means you can retrain the model on images that you want. So once you've got stable diffusion, the open source version on your computer, the first thing you'll do is to remove the couple of lines of code, which exist as the guardrails, which whenever stability, AI is asked about sa diffusion generating illegal content, they will say, oh, we've got these guardrails, blah, blah. And, we are very against the use of our technology to generate illegal content, blah, blah. So I mean, releasing something open source means that you, I believe, you can't be touched legally for what's created using it because. It's almost nothing to do with you, even though you've put it out there. You can download the data sets and retrain the model on what you want, which then means that when you put in your prompts, it will generate images that are closer to what you desire. So you can train it on lots and lots of pornographic images. You can train it on images of real child sexual abuse. Having said that, because of the way AI image generators work, they can actually put two context, two concepts together. So if something understands nudity and pornography and the concept of a child even clothed, then it can put them together. It doesn't need to be fed images of, real child sexual abuse material in order to generate. Okay. That's quite significant cuz a lot of people won't understand how this, and this is making things very clear. Very clear indeed. So you, one's preference is the selection criteria, isn't it? Whether you're going for young, whether you're going for a specific, colored hair, a specific shaped body, literally your preferences can be designed into by merging several images of things that you like. Yeah. Well, you're not really Yeah. Mer yeah. Merging the images, but yeah you're training it on its own understanding of the concept of breasts or women or whatever. Yeah, this is a number that I'm just making up myself, but I wouldn't be surprised if maybe 99% of stable diffusion generated images are pornographic. I mean, if you go onto Instagram and you type in hashtag stable diffusion, it comes up with over a million images and scrolling through, you'll see that most of them are of, east Asian young women with huge breasts and tiny waists, in bikinis or whatever.

Octavia 52:44

So I suppose by pornographic, I don't mean that in the normal explicit sense of the word, but sexualized images of young women and girls clothed or unclothed generated by men. For men, it seems explicit or not, it certainly feels like 99% of AI generated images do fall into this category.

Yoyo 53:10

If you go on to discussions on Reddit or other forums, sometimes people will ask, why are people only using stable diffusion for this? And people just reply, look, pornography has driven every technology we've ever had. The internet, I mean, home video, DVDs, it's all driven by pornography. So this is the same. What you are saying really is fairly complex when you think that the general population won't understand it. So we're talking about a minority of people, aren't we, that are in this club? From what it sounds like. It doesn't sound like something that's being created by mass population, but may be viewed by a larger population, but generated and built and designed by a maybe a smaller group of people. Cause it isn't straightforward what you are talking about. Well, the thing is it's getting easier and easier. And more accessible. Yeah. It's getting easier and more accessible. I mean, definitely if you are already quite tech savvy then it's much easier. But actually no, I haven't tried stable diffusion myself, partly because I don't want to crash my laptop.. But I mean the fact that it was trained on initially something like 6 billion text image pairs that then can be, that then exist in such a tiny 6 billion, they cut almost no space at all. Yeah. I would say they are being made. It's not a case of being made by fewer people and consumed by more. It's being made, they're being made by a lot of people. But,, it does take some effort to generate the images you want, because you have to use the words if you want to generate a very specific image, prompt, craft as it's known, which is essentially using the right words and waiting image. So, often you can use a negative prompt. So old ringly. Whatever you'll put all in the negative. To your point around prompt craft, I noticed also that they're using wrapped instead of raped now. Do you know what I mean? They're misspelling words so that they can get those prompts in. Actually I think that once you've got stable effusion offline and you've got your own, you've got the open source version and you've taken away the guardrails, you don't need to do that. But yeah, it's true. People are using, where there are bands onwards, then people do that and it will adapt to that fairly easily and learn what you mean by wrapped. obviously my story is not about pornography, it's about child sexual abuse material, which of course,

Octavia 56:07

wherever pornography goes, this stuff follows or is a part of it. So with the sexualized images of women and girls that are being made, a sub by stable effusion, a subset of that is images of children and actually,

Yoyo 56:25

a related concern is that the images of supposedly adult women, the faces look very young. So you'll often see what looks like a child's face on, an adult woman's body. And the body won't be realistic at all. The breast will just be, beyond outta proportion, the waist so tiny. And yeah, the face so young. And with that, I think there'll be no way to prosecute images like that, even though. Images of children in the UK that are artificially generated are treated the same as images of real children legally and are illegal. But I think there will be a lot of images which can't be prosecuted because the people who've generated them will say, oh, well some over 18 year olds have very young faces and this is what that is. So you'll often see, a disclaimer saying all these people are over 18, even though they don't look over 18. But the images I was looking at are very clearly, pre pubescent children and some even babies and toddlers. So yeah, it's not easy. It's not easy to see that even for the purpose of the research you are doing, Octavia. It's incredibly unpleasant actually, isn't it? Yeah, it was absolutely horrible to see and I. Yeah I think I had this idea that people who had a sexual interest in children that, pedophiles in general, or for the most part, were kind of constantly battling with their consciences and didn't really want to harm children and, saw themselves as ill. And actually with these images, that is not the case. Or at least, the people making the images will often tag them as rape or baby rape or whatever. And the people in the comments to talking about wanting to harm real children and using horrible language. So, whoever's kind of consuming or purchasing those images, they are at least happy with the title being that. the people consuming these images are very bad, horrible people. So definitely I thought, wow, there are actually people out there writing, it's a child's fault when we abduct them for looking like such a slut or whatever. And just these, yeah awful comments and yep, I was, very disturbed and I did struggle, but at the same time I thought, well actually this isn't something that I'm going to just go down this rabbit hole and then be in this kind of pit of despair and not able to do anything about it. I thought, actually, yeah, this is a big story that hasn't really come out yet. I think I can actually make a difference. Not to, I can't stop stable diffusion, but I think if I can raise awareness about the fact that. These images are being bought and sold on the surface web, uh, and are creating communities for pedophiles to talk about their desires to harm real children, then actually, I can make a difference and I can you sort of take to task those companies that are facilitating this basically. Question for you. There is an argument, because you stated this in your report, that some people argue that artificial child sexual abuse material actually prevents real harms to real children. Now what you've just said there is that actually in the way that these individuals are communicating with each other about viewing the images and then conveying and communicating their intent would completely diminish that argument, wouldn't it? Yeah. this is something I've sent you some writing about, but I haven't published anything about this yet. And that's something I'm definitely working on because it seems to be the kind of go-to response for a lot of people. An intuitive thought of, oh, okay, so there's now a material out there of, children being sexually abused that isn't of real children. Of course, that's putting aside the fact that often you can use an image of a real child as a specific prompt and generate the likeness of a specific child. But putting that aside, if it's not an identifiable child and also putting aside the argument or the fact that police and law enforcement are very concerned about the fact they won't be able to identify real children and keep real children safe because they won't know whether a child. Is real or not. Putting aside all of that. People will say, and actually the comments on the Instagram reel that was made by the b C about the piece I worked on, uh, on this subject, a lot of the comments are discussing this and saying, it's so awful, it's so grotesque and evil, but I would rather it be of not real children if these images are gonna be made, it keeps, it could potentially keep real children safe. And, there are these kind of people will then debate that and say, oh, but you know, the way dopamine works is you want more and more. And of course it's gonna harm real children. So it's definitely, it's a response that people have and it's not just people who, it's people who are very concerned about protecting children. Wonder if this might, In some perverse way be a good thing. But actually, yeah, that's not the case. The current conversations are that they're now being able to show a link between the escalation of watching fairly vanilla porn to then being over periods of time needing to have greater satisfaction, and then moving into, extremes that one wouldn't necessarily think they would to begin with. And so they're now identifying trends that very quickly people are going into, b d sm, choking, strangling, using force and violence and things like that. And I'm gonna ask you this question, you see, because there is a SEC security community that is the listenership for the Security Circle podcast, and I have to ask you that these ultimate harms to the mind. And the continuing growth of harms to the mind. They disenfranchise people, they individuals become disconnected, communities become disconnected. This is often we've identified now considered as a weapon for hostile state actors in the sense of, building nations of just generally disenfranchised as connected people. Something to consider, isn't it? Yeah, no, definitely. Yeah. I think it's not a static thing when these images are consumed. It's part of a process and yeah. What we know about the brain and porn is it's, your preferences do shift and become more extreme and you need more and more extreme content. And I think with this, And child sexual abuse material. There's no reason why material created for sexual gratification when that material is of children. There's no reason why that would be any different. But actually what I found is almost worse than that. In a way, the discussions that go on between the consumers and the sellers and the buyers of this material, the conversations that go on it suggests that actually this is an additional thing, to real child sexual abuse material. It's not an alternative. It has its own benefits basically. And those benefits include, firstly, the fact that it's legal in some countries such as Japan. So what that means is you can have. Millions of these images existing on the surface web, and suddenly that's a place for pedophiles to gather and discuss their sexual interests in children in a really easy way. Where in the past, you wouldn't have those kind of discussions in what is essentially in public, online. so the comments under the images are all, or often people discussing what they want to do to real children that they know. So yeah, I'll probably say a few disturbing things here, but No, it's fine because I think you couldn't, on, the BBC has incredibly strict limitations on just how graphic you can be, but obviously we can. Do that here. I think after you told me some of these comments, it kind of, you have to be so careful in life that you don't actually go around thinking everyone's like this. You have to be very real to the fact that these people exist. They are in our communities. They are not maybe that far away from us as we would like to think, but we can't think that all men have this propensity. And there are some fabulously great men out there that have daughters and nieces and sisters, and they would find what we are going to say utterly heinous. So I just wanted to put that premise in. Yeah definitely. It's such a, it's obviously, it's a difficult one because, I mean, it must be really tough if you are a man and you don't fall into this category because it means that, for example, with childcare, because we know that the vast majority of. Pedophiles are male. It makes sense to exclude all men if you want to really keep your child safe. Basically I think men need to get over it. And I think if men are exposed to some of these comments and they really understand just how bad these people are, they will think, actually, I wanna do more. Yeah. Yeah. And the not all men kind of caption. It's almost, it's like, yeah we know that, but it's almost always meant, so, yeah. Do you know these comments you told me about, I found them ha more harmful than the actual images themselves. Yeah. we already know that most child sexual abuse happens through family members. And I think people forget that. They think, oh, we've got to, keep our children away from bad strangers. And of course it's true that Safeguarding is so key. But one of the comments under the Instagram video of me talking about my research was, my daughter's never gonna go on a sleepover again. And actually people were commenting, well, you know that it's more likely to come from inside your own family. And that's what people need to accept. And that is what these comments show. They're all talking about their own family members, their daughters, their nieces. One for example said, guys, should I rape and kill my newborn niece? And there were many comments responding and discussing this a lot talking about their own daughters and yeah. So it would either be their own family members or the imaginary child in the image because there was a lot of ownership taken over the generated. Images, there's kind of this feeling of ownership from the person who created them. It's seen as if that's their own child. There's a possessiveness, there's an appreciation, a sort of adoration towards the man who kind of made those children. And it's, I think some of us think, well, why would men want to look at pictures of not real children? But actually they, they do because they see them as created and these kind of awful masterpieces. And there's also this huge commissioning element, so people will pay a lot of money and I do want some point to talk about the platform Patreon, where. A lot of this is going on. But yeah, people are paying a lot of money for this. And I think it seems to be about the fact you can commission what you want. You can commission children that look a particular way in particular positions. In particular places, with yeah, particular attributes and facial expressions. And, often there'll be, children will be crying, but sometimes they'll be really happy. And it, people clearly have their own preferences. And another danger is the, just the sheer volume of images that can be created. It's just horrifying. There are millions of images like this out there. One person can make 5,000 images in a day it's just absolutely huge. And because there is an element of. There is some effort that goes into it, this whole prompt craft thing. And that's why I think, PDF files are willing to pay other people to do that because not everyone is going to sit on their computer messing around with prompts. Yeah. And then you have your net and there you have your network of producers. Yeah. And then the finance take me through Patreon then, because I personally hadn't heard about Patreon until, and then until I spoke to you. And then I saw something on Instagram and I was flicking through in a video and someone said, check out my Patreon account if you wanna see more. And I went, oh, that's what Patreon is. Yeah. So the premise of Patreon is it enables creators to fund their work. So basically if you have a podcast like you, or if you are an artist or a comedian or anything really in your, your generating your own content, it enables you to monetize it and to have content that exists behind a paywall on the Patreon site. And you will then have a number of subscribers and you can set up subscriber tiers. So for, $5 a month or $20 a month, and people can, people get increasing rewards for how much money they, pay per month. And the company Patreon, which is a private US company, it's been around for a while and it. Has been valued at 4 billion in the past. Oh, it's a huge, yeah. Silicon Valley company.. So they take between five and 12% of what creators earn. So that's where it differs from sites like Instagram, Twitter, and other places where child sexual abuse is inevitably a problem with Patreon, the purpose of it is selling. When it comes to images, the purpose of it is selling images from behind a paywall. So when you look at it like that, it's obviously going to be the go-to place for people who want to sell artificially generated child sexual abuse material. And actually, with Patreon, I mean, of course that content is technically banned, although I found that they clearly don't have enough. Measures in place they're obviously not scanning images or they're not, their scanning methods are not effective. If patrons selling commodities of illicit material, I would say artificial child sexual abuse material surely there's an audit trail cuz people are putting up credit card details, right? Aren't people just a little bit, nervous about providing their credit card details, the audit trail can catch them? Yeah. I mean it could if people were trying to catch them. And therein lies the problem. All I can say is that I know that, there are a lot of people selling artificial child sexual abuse material on Patreon. When the

Octavia 1:13:08

B

Yoyo 1:13:08

B C sent Patreon a page that I had identified and bearing in mind that page, I could not see the images on the page. They were behind a big padlock sign. But I suspected very strongly because I could see on this Japanese site, Pixiv, which is used as almost a proxy search function for Patreon. It's a sort of public gallery that images will be uploaded to, and it will then say for more stuff and for the uncensored stuff, here's my Patreon, subscribe here. So I then go onto the Patreon and you can see, okay, it's behind a paywall, but. This person is, and you know it's got X amount of subscribers. Yeah, it's very likely that's what it is. So V B C sent Patreon one site and Patreon immediately deleted the page and said, yes, this is semi realistic, and yes, this violates our policies. And they said that. Right. Thanks for letting us know. Yeah. We'll remove it straight away. Yeah. So essentially, but of course all the other ones that we didn't send Patreon the links to are still up there. Yeah. So, it's not my job to search other sites, find all the pages on Patreon that are selling this content and report them all to Patreon, myself. That's not my job. That's what they should be doing. And until they do that, they are making money from this. They are taking the cut of the sale of these images, which are illegal in America. So it's financially lucrative and it's not in their interest financially to go around sending human interventionists into their system to ban the content they're making money from. I'm not saying that they're not trying, I'm saying that whatever they're doing, it's not succeeding effective. Yeah. So, yeah, I mean, all I know is that, that a lot of those accounts are still up there and there are actually more accounts being made all the time because Pixiv, which I mentioned, which is this Japanese site, which is the world's. Most popular image sharing site is, and it's actually I think the 89th most visited site in the whole world. God. And, but you know, I never even heard of it. Yeah, well, so it's it's just an image sharing site. But Pixiv they are actually, to their credit, I mean, they should be. So it's not a huge amount of credit, but they are aware of this. They are putting up blog posts saying, we're dealing with this problem. They've actually apologized for not. Dealing with it sooner. They have said we are not gonna tolerate any realistic images of child sexual abuse on our platform, even though technically it is legal in Japan. I should say that real child sexual abuse imagery was only made illegal in Japan in 2014 after huge global pressure. So they've said that, and they've also said that selling any AI generated content on Pixiv is now illegal. So basically there's pixiv, which is the image sharing site, and then there's the kind of sister site or, well, it's the same site, but it's a sister to it, which is Pixiv fan box, which is monetization of images that you can have behind a paywall on Pixiv itself. But they've banned the sale of, artificially generated, which is almost all stable fusion images. They've banned the sale of that, and that is actually quite effective. So I've seen a lot of pages in, the past six weeks that have now said, look, I'm keeping my Pixiv account, but I'm not using pixiv fan box anymore. I'm only using Patreon. If you want to subscribe and you want the rest of the content, here's my link to Patreon. And yeah, I've seen accounts which, they've literally been selling images of. Babies being raped, saying, oh, patrons ban my account. I've made another one. Here it is. So it's it's a huge issue. And it's not one that patron have publicly engage with, apart from in their response to the b BBC piece. Take me through what some of those hashtags are because I think most of us, won't be aware of them. And it's really good that we keep an eye out for what we see on mainstream social media where we feel generally quite safe. Remember when, we learned that, p os if a child typed in pos and they were talking to an adult, That meant parent over shoulder and there were lots of little codes and things, a lot of parents didn't know that, what sort of hashtags should we be looking out for that we might see in the mainstream that would indicate giving people a signpost to really elicit content? Well, I will say, and I I don't want of course to direct people to this content,. I am definitely worried about like, listing the hashtags. So, so I will say on Instagram, uh, What I have noticed is that on Instagram you won't have stable diffusion accounts with actual child sexual abuse. But there are stable diffusion accounts dedicated to images of kind of clothed, smiling children. And when I had a look at the followers of those accounts, they all said in their bios, I like the little things lover of the little things in life. And yeah, that, I mean, you confirmed you're an ex detective that, that, by that they mean I'm a peed file. But yes, lot of people won't know that. But it's really good to sort of flag that up in case they see it. I should imagine, uh, by the time everyone gets to this stage, they'll be thinking, Christ, I wanna get all of the pictures of my children off the social media. Let's not worry the death out of people. But I think just locking your accounts to make sure that they don't have public access. Least as the very basic layer. And only have people in your Facebook accounts, for example, whom you know and who you can vouch for. And even on Instagram, I try and keep my account closed. I only put pictures of my cat on it, but I obviously we have a hashtag security circle one. You wouldn't believe the amount of trip I get through that on messaging, mainly from Asian websites trying to get me to view anatomy and I'm thinking, Christ it's rife out there. I really feel the kids look. Parents. I know you're not a psychotherapist, but how can parents tell if the stable diffusion tech is on their laptops, we're talking about the age of 13, 14, 15, 16, 17, 18 year olds, how can they tell, what can they do? How if lots of 13 year olds actually. Are using this? I mean, have, I just have no idea. I mean, it's, from the sounds of it from these discussions where these images exist, they basically seem like adult men. But yeah, I think parents, I don't know about the actual children making the images, but I just will say my reaction was to delete or archive all the images on my Instagram of me. that was because I don't want any deep fake stuff being made of me. Also, yeah, if I have, children will not be putting any pictures of them on the internet. So yeah, I mean, there are. Paige is out there on Instagram dedicated to, for example, armpit fetishes of children. And they've clearly just taken, this is not an AI thing. They've just taken pictures of children's armpits from random parents' Instagram accounts, so children are doing gymnastics or whatever. but you're right. And some pedophiles are very specific. One of the ones that I arrested and processed through the system was an ex-teacher. He'd retired already. You kind of think, roll your eyes, you think. Crikey, he'd gone through his whole 30 year career as a teacher before he was ever arrested for anything pedophile related. And his thing was just boys of a certain age round about 11 years old. Creepy cent looking dirty with mad crying, looking distressed. And it was just like, dude, really? This is what you like. I mean, we're not supposed to form any judgments, but he claimed to love them and adore them and it's very hard sometimes. But look, any parents listening there is obviously special. In fact, you can google.gov uk if you just Google help for parents in relation to online content for children. There are lots and lots of places. You've got young minds, you've got think, you've got internet matters gov.uk. You've got some great advice. You've got the UK Safer Internet Center. And what we'll do is we'll just put some links down here so that people can go and get some more advice. Because I think for me, the worst thing would be. There's adults that don't even know how to put a post on Nextdoor, which is a harmless, very Moy. I love Nextdoor. I absolutely love it. I just, honestly, I thrive on it. It's soy though. It's all just, I mean, yeah, but it's all just about like cats. I mean, lost cats, found cats dogs. Yeah. Litter, dog fouling. The helicopters are, what's it looking for? What's going on down that street? All the police are there and Yeah. But you got, you get a lot of people who are still learning how to use that and they don't know how to put. Post up and you just think there are lots of people out there that even don't understand how to do a proper update on their iPhone or their other smartphone cuz we know that they are, superior. And and also the not so smartphone And, I was a joke. And also there are lots of people that don't understand what their children are looking at online. And I think we can't just turn a blind eye now and say, look, I really don't understand it. I think you have to make that effort to try and understand because sometimes it just starts with what's going on in the boys' bedroom, right? Yeah. And the laws really sort of looking to wrap up the laws also, when you think a lot of people think, and I'm gonna be very honest with you, I think a lot of adults just think, look, it's all on the dark web. It's all rather seedy. I don't even know how to get on the dark web. Lots of people dunno how to get on the dark web. You have to know how to get on the dark web. Actually, it's not on the dark web anymore, is it? It's in, the mainstream. Yeah. Yeah. I did want to read out some of the comments because I think people will, okay, people will have this idea or this desire that, pedophiles overall, they maybe want to get better or they convince themselves that children love them and actually they love children. Yeah. And there are actually people out there that want to, they're talking about raping and killing children. So, just for a little, yeah. Just I think to kind of scare people a bit because I think it is important if people are gonna be protecting their children and also to really understand just how bad. This stuff is, that's being made. I think that is important because it will inform people's attitude to the debate over it. Is it really that much of a problem? And then also in general with the, UK's online safety bill. I think it was after seeing all this stuff that I just thought, what if I trusted whoever it was, if it's the N S P C or whatever, I was like, you know what? Give up all my privacy and saw cameras in my home. Do whatever you want. Like, I just don't care. All I care about is protecting children. Like I was so horrified by this stuff. I just think this debate, like we have this debate over, oh, how we, how can we balance our privacy with protecting children online? There is no balance here. Just protect the kids. We don't have, we don't even have like a ripe to the internet. We don't really have. I dunno to talk about like, oh, we'll have a right to like extreme privacy online. I mean, I just think wherever that privacy, wherever we get that privacy, so do these criminals. Apps like session, which enables people to create a, well, basically it means you can communicate with other people through a messaging platform without even adding a phone or an email address so it's like, how does that work? Then you look on session, it's, we'll call itself, it's like the ultimate. Secure communication platform. So of course that is good for people in countries with horrible regimes it is good. But at the same time, all of these men are using session. They're all uploading. Here's my session. Oh, and also I haven't said, of course, that they're saying, here's the real content, here's real stuff. Here's real stuff, blah, blah, blah. One of them said, oh, what kind of real content do you have? And they said, oh, I have anything you want. I even have rape videos with lots of crying. So yeah, wherever this content is, people are including the real stuff. so I think this will help people's attitude towards things like the online safety bill which is, being attacked as very draconian. But yeah, I just think protect children, number one, and that's what it's really about. So yeah, this will actually, so I'll just read out one of the, some of these comments. She looks a lot younger than one year. That only makes my cock harder. And the comments are the nonstop rape. This stupid bitch is enduring after a huge cock pedo just inside her and threw her away. Two more hung pedos show up. Someone says, I hope these pedo live in a big community so everyone will rape her every day. I never even thought such a thing was possible. Do you know what I mean? Yeah, I know, but I'm just gonna go on because then like, it's just so bad that I think without this, people don't understand why I've now got the views that I do about protect children lives. Shameless child rape is the hottest thing in the world. Fuck society. I hope families suffer knowing there's men like me and in capitals kids are not safe. The reply to that yes, I'd love to rape and snuff a baby slut. What more can a newborn rape me hope for in life? Someone says, this is how I want my daughter's children to be used after she pops them out. And then starts, should I rape and kill my newborn? Of course, that's just a masturbator rape yes. Kill. No way. That's a future fuck toy. If you kill it, it's gonna rot. And then let's talk semi footage of you with your newborn niece as brutal as possible. Rape and murder the bitch. She deserves it. Make sure her corpse is unrecognizable and then share, links to where they communicate elsewhere. If it goes bad, just make another one. Do any of you have session? I do. It's, yeah. It's like, it's so extreme and it's so violent I never thought that people would talk, just say, I want to rate them. Yeah. So anyway, everything I just said, like, I think that if you can include maybe like a warning and then just some of that, because without the context and those words, people don't understand the gravity of the situation. Do not understand what, how bad the situation is. You look at general pop prisons, sexual offenders are not popular prisons. No. There's, I think like there's a code, there's a code of offending, right. Bank robbers, whatever. And there's a code, like even in prison of men cannot tolerate men who beat women. So reading those comments. So I'm not a violent person, but I was just filled with, and it actually scared me how violent I felt. I just thought, I want to to kind of calm myself down. I just imagine them, all those men who are, making or buying these images, I just imagine them as really tiny on the floor of the room I was in, and I just like jumped up and down and imagining myself crushing them because it's so. It's, and so, so reading those comments and then you think back to, I'm mad Moak, creator of this technology, saying, well, the net benefit and let's take utilitarian approach. And, people who are worried about AI ethics, they see people as generally bad, but I think they're good. And I think clearly you don't understand, no. As we've learned in human society, even with the greatest technology that we've built, like the Titanic and how we thought we were so magnificently invincible, and you look at how, zebra, the door wasn't shut and now we put alarms in. You look at all of these very significant failings that have led to the cost of life. These are very minute in relation to the cost of life. This could be perpetuating because at the moment there just doesn't seem to be an end in sight. There doesn't seem to be a, an overt society committed to shutting down relentlessly this type of communication. And at the moment, the laws just don't seem to be in line with how fast technology is developing. In the UK is like, we actually have very strict laws, like even a sketch of, child sexual abuse. So just a hand drawing is illegal. People have gone to prison for, child anime stuff in the uk. And with the online safety bill going through Parliament very slowly, I felt quite proud actually to be a citizen of a country that is really trying to prioritize, the protection of children. and seeing that as like non-negotiable rather than write privacy first and then. Yeah. Beyond that, we protect children as much as we can. I just think, when we if we pursue maximum privacy online, children suffer. So how do we negotiate that? How do we manage that? There is no way for us to have that complete anonymity online and also look after children.

Octavia 1:31:00

Briefly one related thing is Facebook saying, oh, we're now gonna have end-to-end encryption on our communication platforms.

Yoyo 1:31:10

Messenger. Oh really? Like WhatsApp's got encryption, hasn't it? Yeah. So Facebook haven't actually had that. And so, great. What that means is that actually I think two thirds of the. Pedophiles or people, trying to traffic children and harm children through Facebook, they are identified through Facebook then, providing law enforcement with private Facebook conversations. And that currently is how, children are protected.

Octavia 1:31:45

And so Facebook then said, right, we're gonna introduce end-to-end encryption.

Yoyo 1:31:49

and that means that those tens of thousands of children would not be able to be protected. And those men wouldn't go to prison or those people? Mostly men. Yeah. And so the UK actually put out, I think some organizations put out an open letter,

Octavia 1:32:05

Saying, look, we understand about the desire for privacy, but if you're gonna introduce end-to-end encryption, then you need to give us an alternative for how we're gonna catch child predators, because otherwise we've got nothing.

Yoyo 1:32:17

And actually, Again, it's a bit like with open source, it's the governments that are then getting a huge amount of criticism for being draconian, et cetera, et cetera. But I think I agree with what they're saying. How do we protect children in those environments? If we pursue this sort of libertarian or just complete freedom online, then children are suffering. And yeah, I think that needs to be acknowledged and yeah not for us not to bury our heads in the sand about just how many people are out there that pose a risk to children and how many people benefit when we release stuff open source, and when we seek ultimate online privacy and anonymity. Well, I hope that we have, stimulated the conversation enough. So that we compel adults to learn more about harmful content, their children's safety, and hopefully lead in, and maybe we might even have people who will step up and champion a fight against this type of really or awful artificial child sex abuse material. So there are people, I think we can, we'll include links to, for example, the Five Rights Foundation. Yeah. Which led by Baras Kidron, who's championing the online safety bill. and yeah, there are definitely people who are aware of this and pursuing laws. There are people out there who are really aware of this. Barness Kidron is one of them. Yeah. We'll put a link to her work, I think.

Octavia 1:33:56

And of course there are lots of other people as well, like Angus Crawford and Tony Smith from the BBC who made that story happen and investigated that with me. So yeah, I'm really grateful to them.

Yoyo 1:34:11

Octavia, all I can say is thank you so much for dedicating the last few months of your life. studying and researching and going quite deep into some really,, unpleasant areas of our human existence. You've done such a great job. You came across incredibly well on the bbc. We'll put your BBC link as well on your bio cuz it's a great story. You own it. And thank you so much, so, so much for your time today. Great. Thank you. It was, yeah, really lovely to speak and really lovely to meet you yesterday.