Radio LogicGenAI is Attacking Trust

Anders Askasen:

Welcome to RadioLogic, the monthly podcast where identity magically makes a difference. I’m joined today by Matt. Matt is from Unrelated. Matt, tell me about yourself and who’s Unrelated?

Matt Harris:

Me, Matt Harris and Toby Harris, and we are unrelated, even though we have the same surname. So we kind of came together with a sort of production background, but then we’ve kind of merged into into using AI over the last mean, I’ve spent sort of eight the last eighteen months kind of really getting into AI, generative AI stuff.

Anders Askasen:

Obviously, that’s why why I wanted to talk to you on the podcast because, as you probably know as well, social engineering is one of these really big hurdles out there. It’s it’s one of the main attack vectors. You the attackers leverage trust to gain advantage and to penetrate defenses and to, you know, trick people, basically exploit trust.

And with the recent developments when it comes to generative AI, and and and I’ve I’ve seen the results. You can easily clone a voice. You can easily clone a video. Where do we stand there? What’s the current situation?

Matt Harris:

Well, when I started eighteen months ago, it was it was just sort of with image imagery. I mean, my background is kind of filmmaking and photography, just working around, you know, with images. And then I I started using Midjourney. It was great, the stuff I could create, but it wasn’t great at replicating faces.

And then I started doing more sort of avatar creation stuff, and it was, it was, it was when I sort of started working with you guys, with Radiant Logic that I thought, actually, wonder I wonder how we can kind of incorporate sort of corporate production using real people, but not, you know, we were filming, we did some filming with you guys, and I thought, I wonder what we can do with AI if we can create realistic versions of yourselves. And it was that start, just the imagery, looking at faces and how you could kind of create avatars from that was the beginning. And And then, obviously, from there, we then went into the voices and using different AIs, and it was, I mean, it was, was, it’s just mindblowing.

I mean, eighteen months down the line, and it’s just getting better and better.

Anders Askasen:

And and that’s kinda scary in itself. Something that pops to my mind is back in twenty twenty four, late twenty twenty four, there was this incident where, I believe it was a Zoom call or something similar, where the actual other party was a deepfake acting as as CFO, and they managed to trick an employee to authorize an a transaction of twenty five million dollars.

And, obviously, that proves that this this could be a dangerous technology.

But we’re are we are we at that point where there’s still tell time tell signs? I know, you know, when we started playing out with Midjourney, which is a generative tool to produce images, there was, like, six fingers and and stuff like that. Is that is that gone, or is that still there?

Matt Harris:

I think that’s gone. I mean, I haven’t used Midjourney. I’ve I’ve used Midjourney a little bit recently, but now that you’ve got the nano banana from Google for for image generation, and it’s incredible. I mean, it’s sort of photorealistic up to, you know, four k easy. Think I think they’re doing eight k now. So, you know, with the even faces you get, you know, the sort of grain, everything, the detail in the face is unbelievable.

Mid journey, maybe, I haven’t really gone into it too much.

I mean, yeah, it, yeah, we’re way beyond the six finger thing. And now, I mean, yesterday or two days ago, the new version of Cling, which is the video generation tool, so the stuff I’ve done with you guys, the Radiant Logic, I could create all the images with Nano Banana, but then the video is made with Cling, so it takes those images, turns it into a moving image.

And now the new version three, three point one or something has just come out, which is even better than the one three days ago.

Anders Askasen:

So so walk me through. Let let’s put ourselves in an adversarial point of view and look through that lens. Right? We we have there there’s plenty of of impressions, videos of someone within a company. Videos on YouTube, you can capture the voice. Walk me through the steps. What’s necessary to capture in order to create something that’s credible and that can fool and trick people?

Matt Harris:

So so in in terms of, like, the stuff of of the films I’ve kind of made with you, you guys, that kind of thing you’re talking about, how we would do that.

I mean I mean, it’s it’s it’s so easy. It it it’s it’s frightening, really, how quickly you can do it. I think when I looked at it eighteen months ago, was, you know, you needed maybe fifteen minutes of audio or something. Now you use eleven labs, take, I think a minute, maybe a minute’s worth of audio, Maybe not even that.

And from that you can clone the voice. So I can pull that audio from anywhere. I don’t need you to say specific things. Can kind of find some audio online, and then it will just clone your voice from that.

And then the facial stuff as as well, you know, three or four photos.

Anders Askasen:

Do you need different angles or

Matt Harris:

Not really, no. I mean, yeah, the more

Anders Askasen:

That will help.

Matt Harris:

It will help. The more you have, the better, but I mean, everyone has got photos of themselves online, right, from all sources, so if you’ve got ten, then, you know, it’s it’s incredible.

Anders Askasen:

So so if if we put this again in an adversarial point of view, I’m an attacker. I I wanna impersonate you, and I go out on social media. I find a whole bunch of photos with, you know I see your face from different angles.

You maybe put out some stuff, your voice in YouTube videos or whatnot. Is that that is sufficient for me to essentially today make an impression of you using those assets.

Matt Harris:

Yeah.

Anders Askasen:

That’s that’s pretty scary.

Matt Harris:

That’s pretty scary. I think I mean, when when I started doing when I actually, when I started doing doing the the the Radiant Logic kind of film stuff, I was in my journey and I was kinda trying to replicate, I think it was you, was trying to create your face, and it just kept it kept throwing up errors. I and I and then I tried other sort of famous people as well, and like, and it just wouldn’t let you do it. So all you do, you just use one of the Chinese AIs, which have no guardrails on them, and it would just let you do anything.

Anders Askasen:

And are they as good?

Matt Harris:

Yeah, well, Cling, Cling is the video one, which is probably the best one out there. And the, well, the the Nano Banana, actually Nano Banana is the Google one, that let me, that that lets me create but I don’t know why. So some have these guardrails and some don’t, which is interesting.

Anders Askasen:

So have you managed to figure out where those guardrails are? No.

Matt Harris:

Not not not really.

Anders Askasen:

No.

So it’s bit of a mystery at

Anders Askasen:

It is. But it is from your from your point of view, it could be a very dangerous tool that, you know, an attacker could use to conduct social engineering campaigns to trick people into giving up, you know, the keys to the kingdom.

Matt Harris:

Yeah, I mean, I think that, you know, the fact that a lot of the American AI tools, oh, we’re gonna build this in, we’re gonna build this in, but you just go and use a Chinese one. So there is no, you know, there is no protection, because they’re not building that stuff in.

Anders Askasen:

So how do we protect ourselves?

Matt Harris:

That’s your job.

Anders Askasen:

So we kind of rely on the American LLMs or AI Gen tools to have the guardrails, but then there’s the Chinese ones or other tools out there that doesn’t have it, and you can’t protect yourself. That’s basically what you’re saying.

Matt Harris:

Oh, yeah. And because I mean, the access, you know, the access is the same. I just it’s just you know, I go online. I use that one instead of that one.

Anders Askasen:

So, I mean, what you’re saying is that we we we actually have something really scary here, that you can build something up fairly quickly. What what what’s the time frame? How do you let let’s assume you you pick some YouTube clips of of myself and you find my voice and you find some social media profile and you get a couple of pictures. Those are, like, vacation pictures, so they’re not they’re not, you know, proper facial shots, if if if you will.

How much time do you need to, you know, create something

Matt Harris:

to?

An hour?

Anders Askasen:

So within one hour, you can have an avatar of me based on whatever I put out on social media with my voice because you found a minute clip on YouTube. Yeah.

With that, you can create something that’s credible, and you can use that potentially to trick someone to

Matt Harris:

do mean, yeah, I could probably find everything, you know, I could say, you know, your Facebook page, your Instagram, say, you know, and sort of go in AI and say, right, you know, create a script based on exactly how this guy writes, and it would look at, you know, your language and your tone and everything, and so I’ve done that with myself as a test to see how that would work.

So then, you’ve AI to write it, then you create the voice clone, and yeah.

Anders Askasen:

So cloning identities is is really easy.

Matt Harris:

In in, well, definitely in terms of what I use it for, to create these films, yes. In terms of, you know, how you would fool someone into bank, I don’t know. I don’t know.

But I would imagine it’s it’s not

Anders Askasen:

a I mean, obviously, you’re not on the bad guy’s side.

You’re on the good guy’s side. So you don’t you don’t necessarily think about how can I use this for malicious purposes? But given the fact that you can do this so easily, there will be people that will use this for that particular reason.

Matt Harris:

Yeah. Yeah. Yeah. I well, yeah, I would think so. I mean, I’m kind of looking when I’m creating stuff. When I’ve created like the films for you, I’m kind of looking at the subtleties of stuff that’s going on and seeing is that, you know, knowing you from down the pub, is that the sort of thing that you would do and and kinda looking at that?

So, yeah, if I can see that, then then, yeah, I’m sure.

Anders Askasen:

Are there any telltale in telesigns today where we are in terms of the the degenerative clones, if you will, that will make it away, like, suspicious, the red flags?

Matt Harris:

I don’t I don’t

Anders Askasen:

I mean, a year ago, it was six fingers, right?

Matt Harris:

Sure. Yeah. No. No. I don’t think so. I mean, we we were just talking I was just showing the guys out there, we were talking about the the Agent Smith film that we did for you guys.

And there’s one scene in it where the Agent Smith walks through the wall and then he stops at the table and there’s a plate of the cookies. And he just looks, he sort of glances down and then looks back. And that was completely unprompted. And I watched that.

And if you look, can kind of see his eyes kind of glance down behind his glasses.

And that I had to, as I stopped and I was telling these guys I stopped, I had to walk out the room because this is just too,

Anders Askasen:

It’s creepy.

Matt Harris:

Too creepy because it’s it’s realized that how this character would act in this scene. And and I didn’t put any of that in. And so that is really frightening.

Anders Askasen:

There’s incidents I’ve heard, I think it’s from the US, where they call up grandparents, and they say that they’ve been incarcerated or put in jail for whatever reason, and they need to pay some kind of bail money. Is is there a mechanism that could be put in place to mitigate that?

We’re talking about people of of age

Matt Harris:

Sure.

Anders Askasen:

That don’t necessarily understand or are aware of the risks of this. Sure.

Matt Harris:

I don’t know, I don’t think so. I don’t think so. I mean, I mean, yeah, my mother is eighty one and I’m dealing with her and trying to figure out apps at the moment. I mean, cannot speak to anyone on the phone because you can’t get phone numbers to talk to companies anymore. You ring them up and they say, no, you need to use the app. And then she can’t use an app. So, and my father was scammed by one of those calls a few years ago.

So yeah, I don’t know, I don’t know how you’re gonna mitigate against that. I don’t know. I mean, that’s a radiant logic job. Right?

Anders Askasen:

I I guess it is. I mean, obviously, this this is a bit of a different take on identity, but it actually is part of what a CISO needs to think about on a daily basis because, social engineering attacks are probably those that will open up the doors and cause some real damage. And here we have a set of technologies that allows you to impersonate someone and use that for malicious purposes.

And it’s, you know, based on what you told me, it takes an hour, and it takes a bunch of clips and some photos, and you’re there.

Matt Harris:

But also, I mean, what you’ve just said as well, like, you know, if you’re a son or a daughter and you’re, you know, kind of ringing your mother, then you’ll be on social media, so everything you’ve written and typed is on social media. So, you know, my mother might read my posts and she would know, oh, that’s how I write. And if you’re a hacker, you can download the history of that person, of you, and get any AI to recreate that language and everything.

So how do you

Anders Askasen:

Have we opened up Pandora’s box?

Are are young people, and I guess it’s not only young people, but people put stuff out there on social media without potentially assessing the risks. Have we opened up Pandora’s box?

Matt Harris:

Yeah. I think so. I think it’s more like the Hellraiser box. Right?

Anders Askasen:

And there’s no way potentially to roll this back.

Matt Harris:

It’s just getting better, isn’t it? Better and quicker. I mean, I said an hour. I reckon it’ll be fifteen, twenty minutes in by next year.

Anders Askasen:

So so you wouldn’t be surprised to see a hacker tool that you just put in a a name, a social media handle, and it will automatically go probe YouTube for some video. Yeah. It will go probe Instagram for some photos, and boom, you have a deepfake that you can put in motion.

Matt Harris:

I’m sure someone’s created an AI agent that will do that already.

Anders Askasen:

Probably. I’m I’m not aware of it, but I I mean, it it sounds like it’s something that probably already exists.

Matt Harris:

Yeah. I don’t know. I don’t know. Unless you, you know, pull the plug on all of this, I don’t know. How are you gonna stop this stuff?

Anders Askasen:

So so the big question is, is there any way we can establish some kind of protocol to have that trust? Like, if I’m talking to what I believe is you on a on a Zoom call, but in reality, it’s not, is is there some kind of mechanism that I can certify that it’s you?

Matt Harris:

I don’t know. Maybe unless they go to the kind of, you know, the bank checking thing where you have your phone and it scans your face, and the only way you’ll be able to communicate will be with video calls in the future.

Anders Askasen:

But we’re not there yet, are

Matt Harris:

we?

No.

Anders Askasen:

I mean, can you trick a system like that?

Like biometric system that

Matt Harris:

I don’t think I mean, that’s why the banks use it, guess, don’t they?

If you do a payment, they wanna see your face Yeah. Until you can do that payment. So maybe it’s gonna incorporate something like that. But there won’t be voice calls, I don’t, because they’re just sound waves, right?

So I don’t know how you can

Anders Askasen:

So back in the days, you had these IVR systems, I think they were called, can basically use your voice to authenticate.

Matt Harris:

Right, right.

Anders Askasen:

I suspect that those are rendered obsolete at this point.

Matt Harris:

Yeah, I would think so.

Yeah. I mean, the voice cloning is is quick and yeah. I mean, it is literally just sound waves, right?

So how

Anders Askasen:

Yeah.

Matt Harris:

How do you secure sound waves? I don’t know.

Anders Askasen:

On on YouTube, there’s been an explosion with these AI generated talking heads that discusses, you know, various things. One of them that comes to mind is is what’s going on with Silver at the moment, and there’s this Asian guy, as he calls himself, John AG.

It’s it’s clearly AI generated, and you can tell it’s it’s a bit clunky, but content is is on on on point. And is that do you think we’re going to see more and more of that type of synthetic content?

Matt Harris:

Yeah. I mean, it’s funny, actually. I watch one. There’s one called Doctor. Julia McCoy, and she’s an AI expert.

And she has she does sort of three or four podcasts a week, and she has a clone, her AI clone. And she said, this is my AI clone. And she wears the Star Trek Doctor. McCoy blue shirt to represent that she’s, you know, this AI avatar.

But she shows you how she did it, how she made this claim. And I watch her because, like you say, the content is great, and I don’t care that it’s not really her. It looks like her. It’s shot in a podcast studio like this, and it’s it’s amazing. And I think, yeah, I mean, people say, oh, you don’t wanna, you know, she’s not real, it’s not really her, but I don’t really see that anymore. When I watch her, I don’t really think that, because the content’s great.

Anders Askasen:

So you don’t care, essentially what you’re Don’t,

Matt Harris:

No.

Anders Askasen:

So so if we’re if we’re wrapping this up with some kind of recommendation to people, should they continue posting, living their social media life, or should they be more conservative?

Matt Harris:

I’d be more conservative. I think I would. That’s me. I’d be much, much more wary of it.

Anders Askasen:

I think with that in mind, we need to think about what we put online.

The moment you put your voice out there, as we’ve heard from Matt, that could be used against you. You can clone that voice and you can create an avatar that can be used to trick them. And the past experiences that you can find in media shows that it’s very easy to trick people, and trust is often the first thing that gets breached. And, that is also part of the wider identity problem. And with that, Matt, thanks for joining the podcast. I really appreciate getting some of insights.