Unveiling the Mechanics of AI Voice Scams: How They Work

0:00Over the last few months, there has been a lot of news around people using AI audio cloning tools to do scams. Essentially, you clone a relative's voice or a person's voice and you use that to scam their relatives. So today on the podcast, we're going to be talking about a new tool that is going to help combat that, as well as what is happening in the entire AI voice scam industry, how those work and how to avoid them. So the new tool that has dropped is the fact that Eleven Labs, which is my personal favorite AI voice sort of platform, they have a really powerful AI voice cloning tool. they have created an AI speech classifier that essentially detects AI generated audio content.

0:42And according to them, it has over 99 % accuracy. So essentially, this new AI speech classifier works by allowing you to upload audio samples, and they scan this for specific unique characteristics. And they're able to tell if this audio file was generated by 11 labs. So if you know, 11 labs was the person that actually generated the file, it can confirm with almost 100 % certainty if the file has not been modified. So I think what is an important caveat to this is that this really only works for content that they have created if you want to get that 99 % accuracy. However, for content that has undergone codec or reverb transformations, this classifier still has a 90 % accuracy.

1:31And they did say that this this like 90 % accuracy, essentially, it drops if the content has been further post processed or additional tracks have been added. But overall, I think this is a really powerful first step towards classifying content that has been created by AI. And since 11 labs is the number one cheapest, but also in my opinion, it's one of the better voice cloning softwares out there. I think that this is a big step for them. They're probably the one that is the most used, most widely available. They have an API. And so I think that they're really making a good stand against kind of some of this malicious use of AI.

2:09Now, the reason that this is important is because there has been an epidemic over the last couple months since a lot of these tools have come out of spam and scam AI voice calls. So typically, what a cyber criminal does is that they, you know, they'll, they'll make a phone call and they'll say that, you know, they're the IRS or they're a healthcare provider and they'll, you know, ask, or, you know, the classic extending your auto warranty. In any case, they have a bunch of different scams and essentially ask people for money. And it's usually older people that get tricked into this. Now they've taken this up a notch now that they have the ability to clone, you know, people's voices.

2:48And in some cases, as little as three seconds of someone's voice can be used to successfully make a clone and they can use that to then go and scam people so mcafee did some research recently and they reported that 77 percent of victims and ai enabled scam calls said they lost money and more than a third of those victims lost over a thousand dollars so this is obviously something that is you know costing people a lot of money. This is something that is proliferating into the scam economy. I think it's important to know that AI enabled voice scams. Essentially, they use these AI voice cloning services and, you know, take a classic scam call a step further and make it seem like it's coming from someone you know.

3:33So essentially how this works is that they target family members or, you you know, and someone that is related to the voice in question. And in the original scam, you know, someone would call pretending to be the police or, you know, call claiming that your friend was in trouble and that, you know, they needed you to send the money for legal troubles or other things. So now the AI scammers are actually just pretending to be someone's child or relative, and they're using these voice clones in the voice of a relative. There's a recent news report, a story about essentially a mom that got a phone call from her daughter saying that she was in trouble and that she needed help.

4:12Luckily, the mom was able to go and contact a relative and ask, you know, what was going on and found out that the daughter was fine at her. I think it was she was at class or somewhere else, you know, that she's supposed to be. And they realized that it was just an AI scammer that was trying to extort them using the voice of their child. So I think that is definitely alarming. I think something else that is bad is that these like a lot of really legitimate tools like 11 labs, for example Also allow scammers to respond in real time As they type out sentences with you know voice cloning apps And so I think some scammers even go as far as to kind of look up different personal information about Who they're trying to scam so that they can help make this, you know phone call more believable And I mean, it's not hard to think of, you know, there's the extreme cases of, you know, someone's daughter calls them crying and is in trouble.

5:04But I feel like there's also like you could make it a lot less, a lot more low key cases where you could say something like, you know, clone someone's voice and, you know, just call their mom and be like, hey, I'm at the gas station and my credit card's declining. I'm not sure what's happening. I'm with somebody here that said if you can Venmo them 20 bucks they could just like put 20 bucks into my account into my you know car for me or like something like that right like just using social engineering and then getting people to just Venmo like a random account that's tied to some scammers you know one of their moles or something like that.

5:36So I think that there's a ton of, you know, really sketchy scams that are going to be very easy to do once people are able to clone voices, now that people are able to clone voices, and I think that a lot of people need to be aware of this. um according to mcafee it's just a very easy to use medium and the attacker doesn't have to have really any expertise in artificial intelligence um in order to pull off these scams i think that at the moment government officials are already seeing signs of generative ai leading to increased voice scams the federal trade commission chair linda khan said earlier this month that um she's seen AI essentially turbocharge fraud and scams, including voice calls.

6:19So this is obviously something that scammers are latching on to. And I think that the McAfee also did a survey and they found that nearly half of adults, so 45 % said in the survey, that they would respond to a voice note or voicemail from a friend or loved one asking for money. And 48 % that they said that they'd send money if they got a note saying a friend or loved one was having, you know, car these AI voice cloning tools that criminals are using, it's tricky because they have a lot of legitimate cases that, you know, make it, you know, you wouldn't want to just outright ban them. For example, I've used them to help me record different content.

6:57A lot of people use them for, you know, like authors using them to help record an audio book or people with speech impediments use them to communicate. And so there's a lot of like legitimate use cases, which make this hard to ban, but it's definitely tricky as a lot of these are being used by criminals. So AI enabled voice scams obviously require quite a bit of effort and not all criminals are going to be willing to make that effort. But I think, you know, for high, you could say high reward, more like large scams, this makes perfect sense. And you definitely are going to see this proliferating into that industry.

7:33I think telecom providers at the moment are also in the last stages of rolling out a really big, um, authentic, like it's a call on authenticity technology, essentially. And it's allegedly going to help flag, uh, robocalls and spam calls as they come in, which of course, if that happens, it's going to be amazing. Um, and so I think there are a couple of things in the meantime that we can do. One of those, a lot of experts suggest that people adopt code words between themselves and loved ones, uh, to use during calls of distress to ensure that they're legitimate. I think that's a great thing. Another thing is to ensure that this is actually the person you think talking on the other end is just to hang up and call them back immediately on their phone number.

8:18And that's an easy way to know if it actually is the person that you are talking to. So all in all, I think that the step by 11 labs to make this new AI speech classifier is a good step, especially since they're one of the bigger platforms doing this. and so obviously you're not going to be able to in live time you know someone's calling you with an emergency run that through an AI speech classifier and check if it's real so I guess that is the one yeah it seems more like something that after the fact you know if someone sent you an audio file maybe you could check or you know the police can use it to check in their investigations but in any case I think using some of the tools like code words calling people back are good steps in order to, you know, overcome this.

9:04Sadly, you know, like, obviously, this isn't the outcome everyone would wish for these AI voice things. But sadly, it's a reality we live in. And we do need to just learn to adapt and move forward at this point. And it's going to be interesting to see what other safety measures are going to be deployed in the future to stop some of these scams, spams, and make overall, you know, phone calls safer into the future.