Ep. 132 - Deepfakes Are No Longer Science Fiction: Fighting AI-Enabled Fraud Together, with Fourthline

Jasmine Chana (0:05)

Welcome back to The Payments Shed. My name is Jasmine Chana, and today, I'm joined by Fleur de Roos from Fourthline, who's the COO there. Fleur, could you start by telling us a bit about yourself and about Fourthline?

Fleur de Roos (0:15)

Of course, Jasmine. Thank you so much for introducing me. So I'm COO at Fourthline. I've been there now for almost five years, and what we do is identity verification, but then, of course, the digital way.

So we onboard retail banking clients from all over Europe, millions a year.

Jasmine Chana (0:34)

Wow. Fleur, how big is the deepfake problem right now?

Fleur de Roos (0:39)

I must say, Jasmine, it's getting much worse. It's quite scary actually. So we did some research and actually 15 percent of Europeans are being scammed. It's a horrifying number, I must say. We also see, and this is terrifying, that it's so easy to do right now. So actually, it's 4.5 percent more profitable to do an AI scam than it was before, but also the accessibility of it.

It just costs five dollars nowadays to just buy some synthetic data, where first it was thousands and thousands of dollars to buy that, or even just 10K to buy a platform, so a deepfake platform. So I think that's the scary part of it, right? That anyone can just buy it and do its own thing with it. And I think that that's what the interesting thing is, whilst it's out there and everyone thinks it's very sophisticated, it's actually not.

Jasmine Chana (1:35)

Yeah, it sounds like it's becoming a lot more accessible in the market, right? Correct. And as COO, you run the team that reviews flagged fraud cases. What have deepfakes done to their workload?

Fleur de Roos (1:46)

Well, actually what we see is that there's a lot of using actual real documents that have been stolen and then using that to create fake identities.

So, what we see is an increase from, I think around 38 to 55 percent of those. So, big search there where those identity documents are stolen or they're bought, right? Stolen documents, but then actual videos are injected by using AI. So, to create like this fake personae, that's what we see. And we also see that then creating fake documents, that's actually going down from 42 percent to 32 percent. So, actually, that's proven less effective. We see now that the jump is to using the AI to inject these fake videos into the onboarding flow.

Jasmine Chana (2:34)

And what do you think's been the main driver of that change?

Fleur de Roos (2:37)

I think it's actually been the success factor of it, right? So, we just see that they're pushing all these new ways of faking that. And whilst they build on that knowledge, that they use even more sophisticated means. So, it's just kind of dropping what used to work in the past, and then moving on to something that now works, right? So, it's just being created.

Jasmine Chana (2:58)

And for your clients, they want speed and certainty. How do you manage those expectations when, let's be honest, the deepfakes just keep getting better and better?

Fleur de Roos (3:06)

Yeah, I think actually we've passed that phase. And what we've seen, so we've done this research, really interesting. It shows actually that the consumers, they are actually okay with friction, as long as they know that their security is taken care of, right? So when they are educated on the checks that are being done and why they are being done to prevent these sort of AI fraudster attempts, they are also fine with it. So where we assume that they want to onboard really, really quickly and within one minute or less, this is actually what they think about it. So actually, we should ask ourselves, why not more accept that and educate our consumers in that that actually protects them. And then they will most likely accept the fact that it takes a little bit longer.

Jasmine Chana (3:47)

Yeah. And how much longer are we talking typically? It's interesting you say that we're okay actually with friction, or clients are okay with friction. Because I think it's something especially in the payments and fintech world, we're used to wanting to reduce rapidly, right?

Fleur de Roos (4:00)

Yes. Yes. Well, I can give you an example. So what we now do, we're used to handling cases within 35 seconds. So it's real, real quick, right? We don't even need to add much more to that, to just build in those few couple of data points that we would like to have verified.

So it's not like it will cost an hour then, right? It's just a matter of adding seconds, maybe up to a minute, who knows? It's not that it will be the end of the world. It's just that we realize that it will cost a little bit of more time.