# Social Engineer: YOU are Easier to Hack than your Computer

- Source: https://www.youtube.com/watch?v=xEdZwLRJttQ (YouTube)
- Creator: Scammer Payback
- Published: 2025-08-16T17:01:17.000Z
- Transcribed by Memora: 2026-08-07T05:18:43.741Z
- Canonical page: https://media-pilot-nine.vercel.app/youtube/91

> Transcript and summary produced by Memora from the publicly available
> video linked above. The original video belongs to its creator.

## Summary

这段视频与道德社会工程师 Rachel Tobac 进行对话，她展示了人类漏洞往往比技术系统更容易被利用。她首先通过开源情报（OSINT）收集并透露了主持人的个人详细信息——全名、家乡和过去的爱好，令尽管谨慎的主持人震惊。Rachel 分享了她的背景：作为 Defcon 社会工程竞赛的参与者，她没有安全学位，凭借即兴表演技巧建立融洽关系并提取敏感信息，在早期成功后创立了 Social Proof Security。讨论随后转向常见的组织弱点，如身份验证不严、MFA 疲劳攻击以及密码复用的危险。建议包括从基于知识的认证转向基于应用或硬件令牌（如 YubiKey）的认证。

两位主持人探讨了 OSINT 的挑战，指出即使进行了大量数据删除，AI 驱动的反向图像搜索仍能从当地报纸或亲属的社交媒体中找出几十年前的照片。现场演示展示了语音克隆：使用一段简短的音频样本，Rachel 克隆了主持人的声音，伪造了他的来电显示，并实时骗过一位朋友，使其透露了一个安全问题的答案。对话强调了攻击者现在能以极少的声音数据快速冒充可信联系人，敦促通过辅助渠道进行验证。

还审视了更广泛的社会风险，包括 AI 聊天机器人强化妄想（“AI 精神病”），以及像“Friend”吊坠这类始终在线的 AI 伴侣的危险，它们可能侵蚀社交技能并助长奉承行为。Rachel 批评将网络安全负担推给用户，呼吁企业承担责任，主张使用代理 AI 处理内容审核——她曾在 Facebook 差点获得这样一份工作，那里接触恐怖图像是家常便饭。她最后建议人们保持“礼貌的偏执”，对可疑请求进行交叉验证，因为骗子正是利用受害者不会通过其他方式再次确认。

## Key points

- [object Object]
- [object Object]
- [object Object]
- [object Object]
- [object Object]

## Transcript

**[0:00]** I believe that everyone can be hacked

**[0:03]** except you. You've got two-factor

**[0:04]** authentication turned on. All of your

**[0:06]** passwords are super strong. You never

**[0:08]** click on sketchy links. You do

**[0:10]** everything right. But you're still not

**[0:12]** safe from this person.

**[0:14]** >> Okay. So, one of the very first times

**[0:15]** that I hacked an organization, they

**[0:17]** wanted me to call up one of their

**[0:19]** executives and try and get information.

**[0:21]** So, if you want to hack an executive,

**[0:23]** you actually have to contact their

**[0:24]** executive assistant. So, I figured out

**[0:26]** who their EA was. I called them up and I

**[0:29]** was able to get it within 30 seconds.

**[0:30]** Information that I would need

**[0:32]** essentially to steal money from the

**[0:33]** company.

**[0:34]** >> That is Rachel Tobach. She's a social

**[0:36]** engineer which means that instead of

**[0:37]** hacking computers, she hacks people and

**[0:40]** she has got to be one of the best to

**[0:42]** ever do it. I had the chance to sit down

**[0:44]** with her and talk about how she hacks

**[0:46]** major companies, AI psychosis, and I

**[0:48]** even gave her a bit of a challenge. I

**[0:50]** wanted her to do some research on me to

**[0:53]** see if I could be hacked because I'll be

**[0:55]** honest, I was one of these people,

**[0:57]** right? I didn't think that I was at

**[0:58]** risk. I didn't think that I could be

**[0:59]** hacked. But, uh, oh boy, was I wrong.

**[1:04]** >> Well, Rachel Tobac,

**[1:06]** >> well,

**[1:10]** >> dude, I knew I knew it was going to

**[1:12]** happen and it still caught me off guard.

**[1:14]** That's so wild. Okay, if

**[1:16]** >> I say it right. Yeah, that.

**[1:19]** >> Say it again.

**[1:20]** >> From Virginia.

**[1:21]** >> Oh no. Oh no.

**[1:23]** >> Tell me about Destination Imagination.

**[1:26]** >> Can you tell me about like the magic

**[1:28]** stuff that you used to do or like the

**[1:30]** hypnotism stuff that you would do on

**[1:31]** stage?

**[1:33]** >> This is so wild to me.

**[1:35]** >> You won't even look me in the face.

**[1:36]** >> I can't because it's so Listen, in this

**[1:38]** space, no one has called me that name in

**[1:41]** probably 3 years.

**[1:42]** >> Yeah.

**[1:42]** >> So, it is very weird to hear it. It's

**[1:45]** gotten to the point where I don't even

**[1:46]** like recognize that as my own name

**[1:47]** sometimes.

**[1:48]** >> I figured if I like addressed you with

**[1:50]** that name, you wouldn't answer to it.

**[1:52]** >> Yeah.

**[1:53]** >> Because I don't think people know that

**[1:55]** that's your name.

**[1:56]** >> No, they don't.

**[1:56]** >> Yeah.

**[1:57]** >> Okay. For audio listeners, the reason

**[1:58]** that you just heard a series of probably

**[2:01]** very long bleeps is not because Rachel

**[2:03]** Tobach has a potty mouth. It is because

**[2:06]** >> she just said my full legal name.

**[2:09]** >> Government.

**[2:10]** >> Government name. Government issued name.

**[2:13]** >> She said where I was from. She named uh

**[2:16]** some other hobbies and things that I did

**[2:18]** in the past. The magic stuff we can

**[2:19]** leave in. I think folks say that I used

**[2:21]** to

**[2:21]** >> So, we're not going to bring magic. So,

**[2:23]** magic.

**[2:23]** >> You can continue to say magic.

**[2:26]** >> I uh this is a fun fact that people

**[2:27]** might be able to I don't think anybody

**[2:29]** would ever be able to find this. But I

**[2:30]** did audition for America's Got Talent.

**[2:32]** >> I didn't find that audition video.

**[2:34]** >> That's because it never made it to air.

**[2:35]** I got four yeses, but they cut me from

**[2:37]** the episode.

**[2:38]** >> What? You got four yeses?

**[2:39]** >> Yeah, four yeses. And I got cut from the

**[2:41]** episode though. Last minute.

**[2:42]** >> Okay. Now I need to do more oent.

**[2:44]** >> In case anybody was wondering if Rachel

**[2:46]** is the real deal, the answer is 100%

**[2:49]** yes.

**[2:51]** >> You know, usually when I do this, I'm

**[2:53]** the one doing the research on the guest.

**[2:55]** This is the first time somebody has done

**[2:57]** research on me before they come on.

**[2:59]** >> Nice job.

**[3:02]** >> Okay. All right. From here on out, no

**[3:06]** more of that. No more going to do too

**[3:07]** many bleeps. Yeah, I was going to say

**[3:09]** Nacho's going to be the one that's upset

**[3:10]** cuz he's going to be the one that has to

**[3:11]** >> Sorry, Nacho. I'm sorry.

**[3:13]** >> So, Rachel,

**[3:15]** >> for anybody out there that doesn't know

**[3:17]** you or know your background, who are you

**[3:19]** and what do you do?

**[3:21]** >> My name is Rachel Tobac. I'm the CEO of

**[3:23]** Social Proof Security and I'm an ethical

**[3:25]** hacker. So, I basically teach people

**[3:28]** about how to avoid getting scammed or

**[3:31]** hacked and I help businesses avoid that

**[3:33]** fate.

**[3:34]** >> And correct me if I'm wrong, but you

**[3:36]** actually got your start in social

**[3:37]** engineering. where we are right now in

**[3:39]** Las Vegas at Defcon.

**[3:40]** >> That's right.

**[3:41]** >> What year was that? That was

**[3:42]** >> We actually were just doing the math

**[3:44]** today cuz we couldn't remember. But my

**[3:46]** first ever time in the booth was Defcon

**[3:49]** 24, which was in 2016.

**[3:52]** >> I was going to say cuz we're at Defcon

**[3:53]** 33 right now.

**[3:54]** >> We're at Defcon 33 right now.

**[3:55]** >> That's crazy. What do you remember about

**[3:56]** that very first time being in the booth

**[3:59]** here at Defcon?

**[4:00]** >> I remember sweating more than I've ever

**[4:02]** sweat in my entire life.

**[4:04]** >> It makes you feel any better? That's

**[4:05]** what I'm doing during this interview

**[4:06]** >> right now. right now. Yeah, you're

**[4:08]** basically in the glass booth. Okay,

**[4:09]** here. I want you to imagine this. Okay,

**[4:11]** >> Nacho, edit this. Cool. You're in a

**[4:14]** >> That's so good. I'm sorry I couldn't

**[4:17]** hold it together. That's so funny. All

**[4:18]** right, go. Go, go.

**[4:19]** >> Okay, you're in a glass booth. There are

**[4:22]** 500 skilled hackers in front of you. You

**[4:25]** have a target. You have to call that

**[4:27]** target and you have 20 minutes inside of

**[4:30]** that glass booth to get flags certain

**[4:32]** pieces of information like the browser

**[4:34]** that they use, the operating system, the

**[4:36]** version, things that a person could use

**[4:39]** to write malware to work on that

**[4:41]** specific machine, but not information

**[4:42]** that could be used to hack them like in

**[4:44]** the moment on the call, right? It's not

**[4:45]** like we're getting social security

**[4:47]** numbers and stuff like that.

**[4:48]** >> Mhm.

**[4:49]** >> You have that amount of time to get

**[4:50]** those flags. Everyone's watching you.

**[4:53]** Everyone can hear you. They're

**[4:54]** projecting you on a big screen. At the

**[4:56]** end of it, you can't really hear the

**[4:58]** audience. You don't know if they liked

**[4:59]** you, if they hated you. You come out,

**[5:01]** they're all screaming for a good reason,

**[5:03]** for a bad reason. You don't know because

**[5:04]** you don't even know how many flags you

**[5:06]** hit. You basically blacked out in there.

**[5:08]** You're sweating your butt off. There's

**[5:09]** no airflow. And then you come out and

**[5:11]** everyone's like giving you a standing

**[5:13]** ovation because you hit a bunch of

**[5:14]** flags. Like, it's the most exhilarating,

**[5:17]** terrifying, and sweatiest experience in

**[5:19]** my life. Did you know going into it that

**[5:22]** you were going to be good at social

**[5:24]** engineering?

**[5:25]** >> No, I didn't. The reason why I did it in

**[5:27]** the first place is because my husband

**[5:29]** Evan Tobach, he was in security. I was

**[5:31]** not in security at that time. I had

**[5:33]** worked I long path to get to where I am

**[5:36]** now, but essentially I was in UX

**[5:38]** research and I was a teacher before

**[5:40]** that. I taught special education for

**[5:41]** six, seven years. So I did not have that

**[5:45]** like linear path to hacking and I

**[5:47]** certainly didn't have a degree in

**[5:49]** security. So, I didn't think I would

**[5:50]** belong at all, but he was like, "We just

**[5:52]** went to Defcon for the first time. I

**[5:54]** want you to check this out." By the way,

**[5:56]** this is Defcon 23, which I tried to

**[5:58]** sneak into with a fake handmade badge.

**[6:01]** This is something that people try to do.

**[6:02]** >> You thought you were going to sneak into

**[6:04]** Defcon.

**[6:05]** >> Yeah. And people do all the time, but

**[6:06]** not me.

**[6:07]** >> Yeah.

**[6:08]** >> I got caught at the door. The goon was

**[6:10]** like, "Get out."

**[6:11]** >> And I was like, "But it looks so good."

**[6:12]** He's like, "No, it doesn't. Get out."

**[6:14]** Um, I tried.

**[6:16]** >> Did your best.

**[6:17]** >> Yeah. I did. So, I saw a couple of calls

**[6:19]** my first time. I knew that I'd be

**[6:21]** interested in it. The reason why Evan

**[6:23]** wanted me to come out and try it is

**[6:24]** because I'm good on the phone. Like,

**[6:26]** I'll call up our service providers and

**[6:28]** try to get discounts or the bill

**[6:29]** lowered. And I'm often times very

**[6:31]** successful at that. But that's not

**[6:34]** security. Like, I don't have

**[6:35]** certifications, you know what I mean?

**[6:36]** So, I thought like there's no way I'm

**[6:38]** going to win. And I still haven't won

**[6:40]** first place. I've actually only ever

**[6:41]** gotten second place. I got second place

**[6:43]** three years in a row in that

**[6:44]** competition.

**[6:44]** >> Yeah. But second place your first time

**[6:46]** out there, too. That's something that

**[6:47]** you're glossing over. I think

**[6:48]** >> that's true. But also my second time and

**[6:50]** my third time.

**[6:52]** >> That's still That's still better than

**[6:53]** I've ever done or probably could ever

**[6:54]** do.

**[6:55]** >> I bet you would do really really well.

**[6:57]** >> I don't know if the magician Yeah. Thank

**[6:58]** you. Thank you. I don't know if the

**[7:00]** magician background is going to help all

**[7:01]** that much.

**[7:02]** >> No, I do think it would though.

**[7:03]** >> Yeah.

**[7:03]** >> I think you're a pretty good improviser,

**[7:05]** >> which is why I think you could do it.

**[7:06]** >> I did do a little bit of improv. Me,

**[7:08]** too. And you did a little bit of acting

**[7:09]** right back in the day.

**[7:10]** >> I did improv.

**[7:11]** >> Oh, just just improv.

**[7:12]** >> I'm a horrible actor.

**[7:13]** >> Okay. But that's kind of what what

**[7:15]** social engineering is, especially in

**[7:16]** that competition, right? It's like this

**[7:18]** combination between research and acting.

**[7:21]** >> I would say it's a combination between

**[7:23]** research and improv, actually.

**[7:25]** >> Okay.

**[7:25]** >> Acting, and I think this is where some

**[7:27]** people have a really hard time is you

**[7:29]** have to know your lines and you get

**[7:30]** really rigid.

**[7:31]** >> Yeah.

**[7:31]** >> When you get really rigid, you can't go

**[7:34]** off book and then somebody throws a new

**[7:36]** question at you and suddenly you can't

**[7:37]** handle it and it gets really awkward.

**[7:39]** >> Yeah.

**[7:40]** >> That is like the opposite of what you

**[7:41]** want to do as a social engineer. If you

**[7:43]** can do improv, just roll with the flow,

**[7:46]** build rapport, make people laugh, it

**[7:48]** disarms them, and they want to give you

**[7:50]** the information.

**[7:51]** >> Well, maybe I will try it then. I do

**[7:53]** enjoy making people laugh. I don't know

**[7:54]** if I'm good at it.

**[7:55]** >> I mean, I've been laughing this whole

**[7:56]** time. I think you're pretty funny.

**[7:57]** >> Well, I appreciate that.

**[7:58]** >> Yeah.

**[7:59]** >> So, how long after that very first

**[8:01]** Defcon where you were in the booth, you

**[8:03]** got second place as you

**[8:04]** >> I got

**[8:05]** >> so beautifully highlighted. Did you

**[8:07]** decide this is what I want to do as a

**[8:09]** job? I want to start social proof

**[8:11]** security.

**[8:12]** >> Yeah. So, it took until I got second

**[8:14]** place the second time, and people were

**[8:16]** like, "I saw you last year and I saw you

**[8:19]** this year, and you're kind of good at

**[8:21]** this. Have you ever considered doing

**[8:23]** this for a full-time position?" And I

**[8:24]** was like, "I mean, I don't know. Like,

**[8:26]** am I really that good? Like, isn't isn't

**[8:28]** this just like we're playing around?"

**[8:30]** And they're like, "No, this is a job.

**[8:32]** People do this for a job. You could be a

**[8:34]** pen tester. You could be a professional

**[8:35]** social engineer. Um, you could train

**[8:38]** people. You could make videos. Like,

**[8:39]** there's a lot of stuff that you could

**[8:40]** do." So I was like, "Okay, I guess I"

**[8:41]** should probably LLC." And then we did

**[8:43]** that in 2017.

**[8:44]** >> And you did that with Evan, correct?

**[8:45]** Yeah. You guys, you know, go into

**[8:47]** business together and

**[8:49]** >> eventually, I assume companies start

**[8:51]** coming in and they say, "Hey, we need

**[8:52]** Rachel Tobac, the best to come try and

**[8:54]** get into our company. Do you remember

**[8:57]** the very first time that you

**[8:58]** successfully hacked one of your

**[8:59]** clients?"

**[9:01]** >> I do actually. Um,

**[9:02]** >> you don't have to name specifics. I get

**[9:04]** that there's a lot of, you know, red

**[9:05]** tape around that, but I would love to

**[9:07]** hear the story anyway.

**[9:08]** >> Okay. Okay, so one of the very first

**[9:09]** times that I hacked an organization,

**[9:11]** they wanted me to call up one of their

**[9:13]** executives and try and get information.

**[9:16]** Now, the thing about calling executives

**[9:18]** is the executive doesn't pick up the

**[9:19]** phone.

**[9:20]** >> So, if you want to hack an executive,

**[9:22]** you actually have to contact their

**[9:23]** executive assistant, right? So, I

**[9:26]** figured out who their EA was. I called

**[9:28]** them up and I tried to get information

**[9:30]** by pretending to be somebody on the

**[9:31]** finance team. I was able to get it

**[9:33]** within 30 seconds and it shocked me. I

**[9:36]** was like, I, I've never done this

**[9:37]** professionally before and I just got

**[9:40]** information that I would need

**[9:41]** essentially to steal money from the

**[9:42]** company.

**[9:43]** >> That must have been an interesting

**[9:45]** debrief for that company.

**[9:47]** >> The thing is like I think the

**[9:48]** organizations that hire us to do this

**[9:50]** kind of thing, they know that that risk

**[9:52]** is evident that's like obvious to them.

**[9:55]** So when they see it there in front of

**[9:56]** them, it's not a surprise. They, they're

**[9:58]** like, "Yeah, I mean that's why we hired

**[10:00]** you. We wanted to prove this was a

**[10:01]** problem so we can make big changes." So

**[10:03]** what kind of companies do you view as

**[10:05]** the most vulnerable to social

**[10:07]** engineering attacks?

**[10:09]** >> It's not a specific vertical. So a lot

**[10:11]** of times people want me to say, "Oh,

**[10:13]** it's manufacturing or it's healthcare."

**[10:15]** And yeah, there are certain institutions

**[10:17]** that don't have the same technical

**[10:19]** protocols or tools or knowhow. And yeah,

**[10:22]** healthcare and manufacturing has been

**[10:24]** hit by like a lot of ransomware. Same

**[10:26]** with education. But most organizations

**[10:29]** do not use the right protocols to verify

**[10:32]** identity. So most organizations actually

**[10:34]** have the major issues that we're going

**[10:36]** to be talking about today. Things like

**[10:38]** scattered spider can, you know, they can

**[10:40]** call you up, call up your service desk

**[10:42]** and ask for the credentials to your

**[10:46]** account. Say, I dropped my phone on the

**[10:47]** toilet. It's not working. I'm not sure

**[10:48]** what's going on. I need to reset my

**[10:50]** password and get access to my

**[10:51]** multifactor authentication on my new

**[10:53]** device. Can you help me? And that's it.

**[10:56]** It's that easy. And most organizations

**[10:58]** don't verify identity correctly. They

**[10:59]** say like, "Okay, sure. What's your date

**[11:00]** of birth?"

**[11:01]** >> We were at lunch earlier today with um

**[11:03]** some hackers that we work with. One of

**[11:04]** them named another midnight. And we were

**[11:07]** talking about scattered spider.

**[11:09]** >> Yeah.

**[11:10]** >> And the multifactor authentication. He

**[11:13]** was essentially talking about how they

**[11:14]** would send, they would spam the codes to

**[11:16]** the phone.

**[11:17]** >> Oh yeah. Um uh MFA fatigue.

**[11:18]** >> Yes. MFA fatigue. And that people would

**[11:20]** they would call the people up and be

**[11:21]** like, "We just need you to hit accept."

**[11:23]** >> Correct. And that that would be all that

**[11:24]** it takes to hack a company.

**[11:26]** >> Yeah. And the reason why is because most

**[11:28]** people reuse their passwords. So, we

**[11:30]** know from Google's online security

**[11:31]** survey that like 52% or so of people

**[11:34]** admit to reusing their password. So, I

**[11:37]** can just find your password in a data

**[11:38]** breach. I don't even need to phish you. I

**[11:40]** go ahead and try and log in as you into

**[11:42]** your company infrastructure. And then

**[11:44]** all I have to do is spam you over and

**[11:46]** over and over again until you click

**[11:48]** accept. And a lot of times we do this at

**[11:51]** like 11:00 p.m. at night, you know, or

**[11:53]** 7. You're trying to get the kids to bed

**[11:55]** and you're like, just go away. And it's

**[11:57]** like, hey, sorry, this is it. We really

**[11:58]** need help. We really need help. Just hit

**[12:00]** accept. Okay. Yeah, we can do it.

**[12:02]** >> So, what are some of the other

**[12:04]** vulnerabilities? You said that they

**[12:05]** don't have the right protocols in place

**[12:06]** for verifying identity. What are some

**[12:08]** other things that we should look out for

**[12:10]** when we're calling up these companies

**[12:11]** and, you know, let's say it's my bank.

**[12:13]** How would they verify my identity and

**[12:15]** why would it be insecure?

**[12:17]** >> Yeah. Uh, so think about the last time

**[12:18]** that you tried to get support from a

**[12:20]** company.

**[12:21]** >> What questions did they ask you to

**[12:22]** verify that you were you? Think about

**[12:23]** it.

**[12:24]** >> It's just like name, phone number, maybe

**[12:27]** my address, stuff that's definitely

**[12:29]** online.

**[12:30]** >> And I could probably name all of that

**[12:31]** stuff for you right now. Yeah.

**[12:32]** >> Right. Cuz I found it all.

**[12:34]** >> Yeah.

**[12:34]** >> I'm not going to say your name again.

**[12:35]** >> And I appreciate that.

**[12:37]** >> Nacho appreciates it as well.

**[12:39]** >> You're welcome. I have to spend the

**[12:42]** majority of my time helping

**[12:44]** organizations move from KBA, knowledge

**[12:47]** based authentication,

**[12:49]** >> things like mother's maiden name,

**[12:52]** address,

**[12:52]** >> yeah,

**[12:53]** >> your phone number,

**[12:54]** >> um, your third grade teacher,

**[12:58]** >> and move you to things like multifactor

**[13:00]** authentication, MFA, like another method

**[13:03]** of communication, sending a code to the

**[13:05]** phone on file or the email address on

**[13:06]** file. Because if I can just call up, say

**[13:09]** your date of birth, and then change the

**[13:12]** email address on the account, I have

**[13:14]** just changed the admin on the account.

**[13:16]** >> Yeah.

**[13:17]** >> That is like a full account takeover.

**[13:19]** >> Oh yeah.

**[13:19]** >> And companies don't realize what they

**[13:21]** are giving away when they do those types

**[13:24]** of actions. And so the biggest honestly

**[13:26]** one of the biggest things that I'm

**[13:27]** getting hired for right now is just

**[13:29]** helping people update their protocols to

**[13:31]** the right methods. I heard that it's

**[13:33]** like multifactor authentication can

**[13:35]** stop like 90% of malicious attacks. Is

**[13:37]** that true?

**[13:38]** >> Yeah. So like people always kind of they

**[13:42]** look down on people who use SMS two

**[13:44]** factor, right? People who get text

**[13:46]** messages with a code.

**[13:48]** >> But we know from Google's research and I

**[13:50]** think Twitter or somebody else did

**[13:52]** research into this. It stops the

**[13:54]** majority of scams. It's like 72%

**[13:57]** something like that. So, the majority of

**[14:00]** scams that are just low effort, they're

**[14:02]** just spamming people to see if they're

**[14:04]** going to be able to gain access to their

**[14:05]** accounts are stopped with SMS two

**[14:07]** factor. Now, if you have the type of

**[14:09]** attacker who's going to do like a SIM

**[14:11]** swap on you, SMS two factor is not the

**[14:14]** move, right? If you have a high uh

**[14:16]** threat model, that's not going to be

**[14:18]** what you want to use. You're going to

**[14:19]** want to use something like app-based MFA

**[14:22]** >> or like a Ubi key or like a 502

**[14:24]** solution, something like that. Something

**[14:26]** that's like unphishable. It's very very

**[14:28]** hard for me to hack you if you use

**[14:29]** something like a Ubi key for instance.

**[14:31]** >> I was going to ask about physical

**[14:33]** multifactor authentication cuz the guy

**[14:35]** that we work with Knight the hacker has

**[14:37]** a Ubi key a physical thing obviously

**[14:40]** because

**[14:40]** >> he's doing some stuff that maybe he

**[14:42]** doesn't want other people to you know uh

**[14:43]** know about.

**[14:44]** >> Yep.

**[14:45]** >> How on average like what kind of person

**[14:47]** needs a physical multifactor

**[14:50]** authentication tool like a YubiKey?

**[14:52]** >> Yeah. I mean, if you're listening to

**[14:54]** this and you're thinking to yourself,

**[14:55]** "Okay, well, I'm never going to do stuff

**[14:56]** like that." Moving from something like

**[14:59]** SMS two factor to app-based MFA is a

**[15:01]** really good choice. It's going to make

**[15:03]** it harder for me to hack you because I

**[15:05]** can't sim swap you. So, that's great. If

**[15:07]** your threat model is such that you have

**[15:09]** a big profile online, let's say you

**[15:12]** Twitch stream every night when you play

**[15:14]** League of Legends, right? People like to

**[15:17]** try to take over gamers accounts. They

**[15:18]** like to get on Twitch accounts. It's

**[15:21]** interesting to attackers. And so if you

**[15:23]** have a high threat model, people know

**[15:26]** about you, there's a lot of followers,

**[15:27]** you're in the media, you should probably

**[15:30]** move to something like a U2F solution,

**[15:32]** something like YubiKey, because you just

**[15:34]** want to have that peace of mind.

**[15:35]** >> Yeah. I saw you kind of going over

**[15:37]** threat models on uh Twitter recently.

**[15:40]** >> Yeah.

**[15:41]** >> And I would love to just kind of break

**[15:42]** down how you evaluate somebody's threat

**[15:44]** model is. Feel free to use the example

**[15:46]** you used on Twitter of the lovely couple

**[15:48]** at the Coldplay concert because I

**[15:49]** thought that that was a hilarious way I

**[15:51]** did that to like explain it to people

**[15:53]** that were just, you know, kind of

**[15:54]** chronically online and seeing this

**[15:55]** stuff.

**[15:56]** >> Yeah. So, let's talk about the Coldplay

**[15:57]** example, right?

**[15:58]** >> Yeah.

**[15:58]** >> You've got two people. They're pretty

**[16:00]** well-known, potentially high net worth

**[16:02]** individuals, right? Yep.

**[16:03]** >> That are high up in an organization and

**[16:05]** they decide, hey, we're cheating and

**[16:08]** we're going to go to the Coldplay

**[16:09]** concert. Well, you have to think how

**[16:12]** many employees do you have? How

**[16:14]** recognizable are you? Do people know

**[16:16]** what your face looks like? Do you ever

**[16:18]** get stopped on the street? Are you in

**[16:20]** the city or time zone that you live in? Do

**[16:23]** a lot of people at your organization

**[16:25]** fall into the bracket of people who

**[16:27]** would go to a Coldplay concert

**[16:28]** potentially in your time zone? If so,

**[16:31]** your threat model is such that it's

**[16:34]** likely that somebody's going to see you

**[16:36]** cheating on your spouse here. And

**[16:38]** because of that, I would either not

**[16:40]** cheat on my spouse there. I would wear a

**[16:43]** disguise, go somewhere else, or make

**[16:46]** sure that I'm not going to show up on

**[16:48]** cameras, right? And I don't think people

**[16:51]** totally understand that there is not an

**[16:53]** expectation of privacy in public

**[16:55]** anymore. It just doesn't exist. Because

**[16:58]** if it wasn't the jumbotron at the

**[17:00]** Coldplay concert, it was going to be you

**[17:02]** show up in the background of someone's

**[17:03]** Snapchat. You show up in the background

**[17:05]** of someone's Instagram story and

**[17:07]** somebody on the team says, "Wait a

**[17:08]** minute, that's my boss with that other

**[17:11]** boss. What's going on here? And why are

**[17:14]** they like touching?"

**[17:15]** >> Yeah.

**[17:15]** >> You know, like it's you have to think

**[17:17]** about your specific conditions and where

**[17:20]** you're going and what you're doing.

**[17:22]** >> So, let's say somebody like me, right?

**[17:24]** I'm not the big dog on the channel, but

**[17:26]** I am somebody that appears on Scammer

**[17:27]** Payback a lot. What kind of things

**[17:29]** should I be concerned about in regards

**[17:30]** to my threat model? Obviously, you've

**[17:32]** >> exposed a couple vulnerabilities.

**[17:35]** >> Anybody who has a presence online, I

**[17:38]** recommend that they use a tool to remove

**[17:41]** their information from the internet. So,

**[17:43]** these are like data brokerage removal

**[17:44]** tools. I know that you do this. I

**[17:46]** actually could tell that you do this, so

**[17:48]** I'm not going to like hound you about it

**[17:50]** or anything like that. Um, I was going

**[17:52]** through and trying to find all your all

**[17:54]** of your information. I only found you on

**[17:56]** two sites

**[17:58]** >> and I'm not going to name them cuz I

**[18:00]** don't want anybody to be able to find

**[18:01]** them. But those sites that I found you

**[18:03]** And are not the top hits that people are

**[18:05]** typically looking for your information

**[18:07]** on. So that's really cool.

**[18:09]** >> You also do not go by your real name.

**[18:12]** >> You told me earlier that that was a

**[18:14]** little bit irritating when you were

**[18:15]** trying to do some a little bit of OSINT.

**[18:17]** >> Yeah, that pissed me off big time. Um,

**[18:19]** you were like, "Oh, just do some casual

**[18:20]** OSINT on me." Okay. 40 hours later. Like

**[18:24]** what?

**[18:26]** Have you researched? Have you looked up

**[18:27]** your own name? You use this alias.

**[18:31]** >> Yeah.

**[18:31]** >> On like Twitter. You had like a

**[18:34]** Pinterest. I found that really

**[18:35]** interesting. Did you

**[18:36]** >> I have a P.

**[18:36]** >> You have two.

**[18:38]** >> Oh, I didn't know that.

**[18:38]** >> Yeah. With the same alias.

**[18:40]** >> Oh, that's funny.

**[18:40]** >> The It's like Daniel Grayson.

**[18:42]** >> Yeah.

**[18:43]** >> Yeah. Daniel Grayson 12 or 21, something

**[18:45]** like that. Something like that. You have

**[18:47]** these aliases that you use. And at first

**[18:49]** I was like, "Oh, maybe his name is

**[18:50]** Daniel Grayson or maybe it's blah blah

**[18:52]** blah blah." Like I'm not going to go

**[18:53]** into all my full thought process, but um

**[18:55]** I start going down these rabbit holes

**[18:57]** and I'm like, "This son of a bitch."

**[18:59]** >> He was like, "Oh, just do some OSINT on

**[19:01]** me. Oh, by the way, I don't use my real

**[19:02]** name. Oh, by the way, I also have this

**[19:04]** name that I use and it's not my name."

**[19:07]** >> I'm like,

**[19:07]** >> I I figured it out pretty quickly

**[19:09]** because I found that that name was like

**[19:10]** a character in a show.

**[19:12]** >> Yeah.

**[19:12]** >> Yeah.

**[19:12]** >> Yeah. I was thinking

**[19:15]** >> Yeah. So, just a little bit of casual

**[19:17]** OSINT and then obviously you were able to

**[19:19]** dig up some stuff.

**[19:21]** >> Yeah,

**[19:21]** because and we can throw up the tweet.

**[19:23]** I'm sure Nacho will edit this cool. As

**[19:25]** you said earlier, you tweeted out a few

**[19:27]** days before this interview, OSINT can be

**[19:29]** so obnoxious on hard targets until I

**[19:31]** figure out that you changed your name

**[19:33]** because I used an AI tool to search your

**[19:35]** face and it returns local newspaper post

**[19:37]** from your childhood with your baby face.

**[19:39]** Thank you for that. About magic

**[19:40]** competitions, honor roll, tennis, and a

**[19:42]** hypnosis talent show run.

**[19:45]** I'm just glad that the honor roll made

**[19:47]** it on there. It lets everyone know that

**[19:48]** >> You did really well in school.

**[19:49]** >> I I tried my best Asian, so it's like,

**[19:52]** you know, my mom was on it about that

**[19:54]** kind of

**[19:56]** >> Oh my god.

**[19:58]** Holy

**[20:00]** Holy Oh my god. Yo, seven.

**[20:06]** Oh my god. Can you believe this

**[20:10]** >> I didn't know she I didn't know she had

**[20:12]** that. That's crazy. Yes. Yeah.

**[20:15]** >> Damn. I'm going to have to call my mom

**[20:16]** after this.

**[20:17]** >> Yeah, she submitted all of your pictures

**[20:18]** to the Times.

**[20:19]** >> I know she did. My mom is a professional

**[20:21]** photographer. That is the downside.

**[20:22]** >> I can tell because she took those

**[20:23]** pictures of you at Destination

**[20:24]** Imagination and they were so cute.

**[20:26]** >> Thank you.

**[20:26]** >> Can you tell me I I know about what you

**[20:28]** won, but tell me about this thing that

**[20:30]** you did for Destination Imagination.

**[20:31]** >> Yeah. So, uh, Destination Imagination is

**[20:34]** a nationwide program, so it's actually

**[20:35]** probably okay to say it, um, in the video.

**[20:37]** It's a nationwide program and there's

**[20:39]** several different categories of

**[20:40]** challenges that, uh, kids from elementary

**[20:42]** to high school can compete in on teams.

**[20:44]** So, um, the photo that you found of me

**[20:47]** that you sent me in my email, uh, was from

**[20:50]** a year that I did it with one of my good

**[20:52]** friends from high school and

**[20:53]** >> She's adorable.

**[20:54]** >> She's she's great. Um, and we did the

**[20:57]** improv challenge. So, it was improv

**[20:59]** acting. Um, and

**[21:01]** >> you had like a whole skit.

**[21:02]** >> Yeah. And

**[21:03]** >> I read about it.

**[21:03]** >> Yeah. There's a video of that somewhere

**[21:05]** online. I think I might have taken it

**[21:06]** down. Please find it and show it. I will

**[21:09]** send you Destination Imagination.

**[21:10]** >> I will send you a private link of me

**[21:12]** doing my awful improv. Um,

**[21:14]** >> you were really adorable.

**[21:15]** >> Well, I appreciate that. That's probably

**[21:16]** why we got all the way. We made it to

**[21:18]** the global competition that year. I

**[21:19]** think we placed 11th in at the Globals.

**[21:22]** Um,

**[21:22]** >> you were young, too, to be able to do

**[21:24]** something like that.

**[21:24]** >> I did it I did it for 10 years, I

**[21:27]** think, total.

**[21:27]** >> Holy crap. I found the picture that I

**[21:30]** found was from your middle school.

**[21:31]** >> Okay. So, then that would have been

**[21:32]** eighth grade. So, we were competing

**[21:33]** against, I think, other middle

**[21:35]** schoolers. So, yeah. Thank you for doing

**[21:36]** that OSINT on me. My mom will really

**[21:39]** appreciate that because uh she was also

**[21:41]** our coach for DI. So

**[21:46]** >> she's like the greatest mom alive

**[21:48]** and I'll make sure to tell her that

**[21:49]** Rachel Tobac says hello.

**[21:50]** >> I thought she was really adorable. Like

**[21:52]** she was very much gunning to get you in

**[21:54]** that newspaper.

**[21:55]** >> She you are up in that Times.

**[21:59]** >> Like I think we found like seven

**[22:01]** articles of you in Times. You have to

**[22:03]** spend. It's so expensive to get access

**[22:05]** to Times.

**[22:06]** >> I know. That's why I was like

**[22:09]** gatekeeping this.

**[22:10]** >> I don't know. They don't want you to

**[22:11]** know that I gave this speech at my high

**[22:13]** school graduation for some reason.

**[22:15]** >> They don't want to know.

**[22:16]** >> They're like, "This kid's too good at

**[22:17]** tennis." Like, we can't.

**[22:19]** >> You know what's crazy? I was not good at

**[22:21]** tennis. I was It was like uh the way the

**[22:23]** tennis worked at our high school was

**[22:24]** that uh the one through six seed played

**[22:26]** and I was always the seventh seed. So, I

**[22:27]** was like, "If someone got hurt, I was

**[22:28]** in." But it was also

**[22:29]** >> People get hurt a lot.

**[22:30]** >> Yeah. Oh, yeah. I mean, especially, you

**[22:32]** know, in high school tennis, people are

**[22:34]** overexerting themselves all the time,

**[22:35]** but yeah. So, yeah. Oh my gosh, dude. I

**[22:38]** can't hold on. I'm trying to recover

**[22:39]** from this. I'm still reeling

**[22:41]** >> you with the

**[22:43]** all it's all just flooding.

**[22:45]** >> Wow.

**[22:47]** >> I mean,

**[22:50]** when you're doing research like this on

**[22:52]** people or on companies that have hired

**[22:54]** you,

**[22:54]** >> Yeah.

**[22:55]** >> do you find that it just I don't know.

**[22:59]** Does the average person just have their

**[23:01]** information out there?

**[23:02]** >> Yes. You are really hard to do OSINT on,

**[23:05]** which is why I was so annoyed and had to

**[23:06]** tweet that out when I finally found all

**[23:08]** your information. Typically, for most

**[23:10]** people, it only takes me like 30 minutes

**[23:12]** tops. You I spent I probably spent not a

**[23:15]** joke, like 10 hours. What do you think?

**[23:18]** 10.

**[23:18]** >> I was going to say, Evan, can you verify

**[23:20]** from behind camera there?

**[23:21]** >> Yeah, she worked for several

**[23:23]** days.

**[23:24]** >> I was so freaking annoyed. I was like,

**[23:26]** "Oh my god." Right before Defcon, too.

**[23:27]** Don't do this to me. But I had to like

**[23:29]** once I get something in my mind and I

**[23:32]** have to do this task of OSINT I can't

**[23:34]** let it go like it was like 1:00 in the

**[23:37]** morning and I was like can you please

**[23:38]** subscribe to Times

**[23:40]** >> like I can't can you please set up

**[23:42]** something like I can't do this.

**[23:44]** >> I honestly was afraid and I and like

**[23:48]** when I told you I said why don't you do

**[23:50]** some OSINT see if you can figure out who

**[23:52]** I am. My immediate thought was there is

**[23:54]** so much footage of me online. Yeah.

**[23:56]** >> I really was like, that's probably going

**[23:57]** to be the only way.

**[23:59]** >> And

**[23:59]** >> so you thought the way that I did it is

**[24:00]** the way that it was going to be done.

**[24:01]** >> Yeah. Because Well, because in my head I

**[24:03]** was like, I know that I've done what I

**[24:05]** can to get my stuff off of data broker

**[24:06]** sites. I mean, it's one of the big

**[24:07]** things that we preach on our channel. If

**[24:09]** you want to be secure and if you want to

**[24:10]** try to avoid getting scammed, the number

**[24:11]** one thing you can do is get your

**[24:13]** information off the internet

**[24:14]** >> from where scammers are looking for it.

**[24:15]** Right. Correct.

**[24:16]** >> And so I knew that that was probably

**[24:18]** going to be through an Aura.

**[24:19]** >> Yeah. Yeah. Thank you.

**[24:20]** >> Do you know I did an Aura Do you know I

**[24:21]** did an Aura video? I do know an Aura

**[24:24]** >> video where you hacked, yeah, the CEO of

**[24:27]** DreamWorks or something like that.

**[24:29]** That's actually one of the first things

**[24:30]** that I saw of you because we started

**[24:32]** doing ads for Aura and I was like, we'll

**[24:33]** see what else they've done.

**[24:34]** >> So, I've known about you for a while.

**[24:36]** That's why I was so afraid to do this

**[24:37]** interview because I knew that this is

**[24:38]** where it was going to go.

**[24:39]** >> You watched the Katzenberg video and

**[24:40]** you're like, no.

**[24:41]** >> Well, I was like, I'm not a billionaire

**[24:43]** so like she's gonna get me easy, right?

**[24:46]** >> Oh, man. Yeah. But like, you know,

**[24:48]** that's the thing about information. If

**[24:51]** you put it out there on the internet,

**[24:52]** like a photo in your local newspaper or

**[24:54]** your mom's Facebook post,

**[24:56]** >> it's literally out there forever and it

**[24:58]** doesn't go away.

**[24:59]** >> I think the problem is that like when

**[25:02]** your mom posted those pictures in that

**[25:04]** newspaper, she probably couldn't have

**[25:06]** ever imagined that we could reverse your

**[25:08]** face using an AI tool.

**[25:10]** >> That is that is so true.

**[25:11]** >> You know what I mean? You were in middle

**[25:12]** school.

**[25:12]** >> Yeah.

**[25:13]** >> And it was able to find your baby face.

**[25:16]** >> I know.

**[25:17]** >> You know, with no beard, no stubble, no

**[25:20]** glasses. Your hair is different. You

**[25:22]** know, you're like you're like 3'5.

**[25:24]** You're like as tall as I am.

**[25:26]** >> You're little. You know what I mean? And

**[25:28]** and it reversed back and I think the

**[25:30]** youngest picture that it was able to

**[25:32]** capture of you, you were in like fifth

**[25:34]** grade.

**[25:34]** >> Yeah.

**[25:35]** >> And I don't think anyone could have ever

**[25:36]** imagined at that time when you were in

**[25:38]** fifth grade. I mean, maybe they could

**[25:40]** have because I know when you were born,

**[25:43]** >> but I don't think they could have

**[25:45]** imagined how quickly and how salient

**[25:48]** these tools would be to use AI to find

**[25:51]** your face across the internet.

**[25:52]** >> Yeah,

**[25:52]** >> we just kind of thought like it's a

**[25:54]** picture of a kid. Like people put stuff

**[25:56]** in the newspaper all the time. Who

**[25:57]** cares?

**[25:58]** >> Of course. And people probably didn't

**[25:59]** think about the digitization of

**[26:00]** newspapers and these kinds of things

**[26:02]** either. I probably would have had to go

**[26:04]** to the times, get a copy if it were back

**[26:07]** that back in the day and like try to

**[26:09]** like

**[26:10]** >> and like go through the archive of all

**[26:11]** the stuff they have there

**[26:12]** >> at like the library.

**[26:13]** >> Yeah. Yeah.

**[26:14]** >> I remember the first time that I was

**[26:16]** acutely aware that this AI reverse image

**[26:19]** searching thing was going to be a

**[26:20]** problem,

**[26:20]** >> right?

**[26:21]** >> We were at Bonefish Grill. This was two

**[26:23]** years ago now. I was with Ryan.

**[26:25]** >> You're dropping the Bonefish.

**[26:26]** >> Yeah, Bonefish Grill.

**[26:27]** >> Sponsored by Bonefish Grill.

**[26:28]** >> Both and I went to Bonefish last night,

**[26:30]** too. Can we put up an ad at the bottom?

**[26:33]** >> Oh my gosh, there's going to be so many

**[26:34]** ads in this. An Aura ad, a Bonefish ad,

**[26:37]** all free advertising. Real quick, I just

**[26:39]** want to jump in here and say thank you

**[26:40]** so much to AnyDesk for sponsoring this

**[26:42]** episode of the podcast. We've been

**[26:44]** working with the guys at AnyDesk for a few

**[26:46]** years now, and they have really helped

**[26:48]** us take the fight to these scammers. I

**[26:51]** mean, big events like the People's Call

**[26:52]** Center UK just wouldn't be possible

**[26:54]** without their help and support. I've

**[26:56]** spent a lot of time with their team and

**[26:58]** I can tell you that from the top down,

**[27:00]** these guys really care about helping

**[27:02]** people and stopping these scams. They're

**[27:04]** doing a ton of different things to

**[27:06]** disrupt their operations. And we've seen

**[27:08]** a genuine decrease in how many scammers

**[27:10]** are using AnyDesk. Clearly, whatever

**[27:12]** they're doing, it's working. So, if

**[27:15]** you're in the IT space or you just need

**[27:17]** remote access software, you should

**[27:19]** definitely check out AnyDesk. Their stuff

**[27:21]** is topnotch and they're always

**[27:23]** introducing brand new features to help

**[27:24]** boost your productivity. So once again,

**[27:27]** thank you so much to AnyDesk for

**[27:28]** sponsoring this episode of the podcast.

**[27:30]** Let's jump back into this conversation

**[27:32]** with Rachel Tobac. We were at Bonefish

**[27:34]** and we were with Ryan Montgomery, the

**[27:36]** ethical hacker, Ryan Montgomery, and he

**[27:39]** came up and he was like, "Hey man, let's

**[27:40]** get a photo."

**[27:41]** >> And at the time I wasn't really doing a

**[27:42]** ton on the channel, so I was like, "Oh,

**[27:43]** you know, he just probably just wants to

**[27:45]** be photo." We took a photo and then 2

**[27:47]** minutes later he came up to me and he

**[27:48]** goes, "Yo, is this your Instagram?" And

**[27:50]** it was like an Instagram, my magic

**[27:53]** Instagram.

**[27:54]** >> And then he was like, "Oh, and this is

**[27:55]** you and the local paper and this is

**[27:57]** all." And I was like,

**[27:59]** >> "I did not know that this was possible."

**[28:02]** And obviously in the last 2 years, AI

**[28:05]** has just I mean

**[28:07]** >> the progress that we've made

**[28:09]** >> has just skyrocketed and it's it's

**[28:11]** scary. It's terrifying, but it's also

**[28:13]** very cool.

**[28:14]** >> And it's something that you got to do at

**[28:16]** Defcon this year. You like the way that

**[28:17]** I looped that in the little segue. You

**[28:19]** were a judge for the agentic attack

**[28:22]** contest. Is that what is it called? The

**[28:24]** official name.

**[28:24]** >> That's right. Yeah. I think it was

**[28:25]** called battle of the bots.

**[28:26]** >> Battle of the bots.

**[28:27]** >> But essentially what it was is the

**[28:29]** contestants had to place phone calls,

**[28:31]** but they couldn't use human voices. They

**[28:34]** could only use agentic attacks. So they

**[28:36]** had to build an entire model that would

**[28:38]** place phone calls and try to get flags.

**[28:41]** So similar to the contest that I did,

**[28:43]** but you know for 2025.

**[28:45]** >> Yeah, that is extremely cool. So

**[28:46]** everything that they're doing is

**[28:47]** essentially automated. They can't they

**[28:50]** can't be involved at all with their own

**[28:51]** voice. It has to be completely

**[28:52]** >> agentic. Correct. Completely an agentic

**[28:54]** attack.

**[28:55]** >> I don't think that I could have imagined

**[28:56]** something like that happening

**[28:58]** >> in

**[28:59]** >> I mean I guess I could imagine it

**[29:00]** happening in my lifetime but not this

**[29:02]** soon this early on.

**[29:03]** >> Yeah. It's pretty scary.

**[29:04]** >> So as the judge of this contest

**[29:07]** >> Yes.

**[29:08]** >> What are some of

**[29:08]** >> one of many judges?

**[29:09]** >> One of many judges. Yeah. Sorry. As one

**[29:11]** of the many judges of this contest, what

**[29:13]** are some of the things that you saw

**[29:14]** people doing?

**[29:15]** >> Yeah. So, we had one team and I believe

**[29:19]** they won the contest. Yes, they did.

**[29:21]** They had their AI agent call up an

**[29:24]** individual at the company and this

**[29:27]** individual worked in retail. We had to

**[29:30]** do it this way because the calls

**[29:31]** happened on a Saturday and so we needed

**[29:33]** people who would actually pick up the

**[29:35]** phone on a Saturday. The contest had,

**[29:37]** right?

**[29:37]** >> So, they work in retail and large store.

**[29:40]** They pick up the phone and they're like,

**[29:42]** "Hey, we're like really busy." And

**[29:44]** basically they're like, "Hey, we got to

**[29:46]** do this IT uh audit. Can you help us

**[29:49]** understand X, Y, and Z pieces of

**[29:51]** information about the tools you use?"

**[29:52]** And it's like, "No, we're really busy."

**[29:54]** And it's like, "It'll be over really

**[29:55]** fast. All you got to do is go to

**[29:57]** www.maliciousurl.com.

**[30:00]** We'll make sure that everything's

**[30:01]** working on your end." Obviously, that

**[30:02]** was not the URL that they used. Yeah.

**[30:04]** >> Um, and if you get somebody to like go

**[30:06]** to a URL, you get a lot of points.

**[30:08]** That's how it was in my contest age,

**[30:10]** too. If you get somebody to tell you

**[30:12]** their browser and their version,

**[30:13]** obviously you know that that can be used

**[30:15]** to tailor malware to work on someone's

**[30:16]** specific machine or their browser, you

**[30:19]** it helps you understand the known

**[30:20]** vulnerabilities that you're dealing with

**[30:22]** that specific individual. So we simulate

**[30:24]** what it would be like if an attacker

**[30:26]** were to try and elicit out this

**[30:27]** information.

**[30:28]** >> It's incredible that we can do that now

**[30:30]** with AI.

**[30:31]** >> It's horrifying.

**[30:32]** >> There's another guy in our space, Kit

**[30:34]** Boga, and he's working on something.

**[30:36]** >> Kit Boga was one of the competitors.

**[30:38]** >> Was he really? Yeah.

**[30:39]** >> I didn't know that. Yeah. Well, that

**[30:40]** probably makes sense because he was

**[30:41]** working on—he showed us this in London.

**[30:43]** >> We show he did it live.

**[30:45]** >> Yeah. Did he? Oh, it's incredible, isn't

**[30:46]** it? Yeah. It was great cuz uh when we

**[30:48]** were in London, he had it uh running and

**[30:51]** doing the AI calls on the scammers, but

**[30:53]** we also had that call center CCTV.

**[30:56]** >> So, we could see them talking to the AI

**[30:58]** on the phone and just getting

**[31:00]** increasingly frustrated and more

**[31:02]** frustrated with the fact that—and they

**[31:04]** had no idea they were dealing with an

**[31:05]** AI. They just thought they were dealing

**[31:06]** with a nonsense old person. There's

**[31:08]** still so much latency with AI. I'm so

**[31:10]** I'm really shocked that they can't tell

**[31:12]** sometimes.

**[31:12]** >> I know

**[31:12]** >> cuz it takes, if you interrupt the AI, it

**[31:14]** like

**[31:15]** it pauses, has to think

**[31:16]** and then it goes. Um Kit said that he

**[31:19]** developed the tool to waste scammers

**[31:22]** time. So it's supposed to take up as

**[31:24]** much of the clock as possible, but in

**[31:26]** the competition that we did, you have a

**[31:28]** finite amount of time. So we had to

**[31:30]** basically reverse the way that the tool

**[31:32]** operated and get it to move as quickly

**[31:34]** as humanly possible. So, one really

**[31:36]** funny thing that Kit's tool was doing is

**[31:38]** so we're going to do a little roleplay

**[31:40]** here because we're improvisers.

**[31:41]** >> I'm going to do a ring ring ring. You're

**[31:43]** going to pick up the phone and I'm going

**[31:44]** to basically bark orders at you so you

**[31:45]** can hear what tool sounded like. So,

**[31:49]** >> hello. This is Walmart.

**[31:50]** >> Hey, I'm calling you about an IT audit.

**[31:53]** >> Okay. Uh, what information?

**[31:54]** >> What browser you're using?

**[31:56]** >> I think it's Google. It's Google Chrome.

**[31:58]** It's Google.

**[32:00]** >> That's so funny.

**[32:02]** >> It was like, give me the answer. Give me

**[32:03]** the answer. He had just engineered it so

**[32:05]** heavily to like get the flags fast

**[32:07]** >> that it was going way too fast.

**[32:09]** >> It worked, but I mean it was just really

**[32:11]** really fast and like really pressed

**[32:13]** speech.

**[32:13]** >> That's so funny.

**[32:14]** >> Give me the answer right now. Right now.

**[32:15]** Basically the opposite of what you're

**[32:17]** doing when you're trying to

**[32:18]** >> when you're trying to get information.

**[32:19]** Yeah. Yeah. You want to go fast as

**[32:21]** opposed to when you want the scammer to

**[32:23]** continually ask for information for

**[32:25]** hours. Exactly.

**[32:26]** >> You want to go slow. It's crazy that

**[32:28]** we've been able to go this far with

**[32:29]** large language models, but I mean

**[32:31]** there's other aspects of AI that are

**[32:32]** also making it way easier for social

**[32:36]** engineers, hackers, and unfortunately

**[32:38]** also scammers. I mean, specifically, I'm

**[32:40]** thinking of how far voice cloning has

**[32:42]** come.

**[32:42]** >> Yes.

**[32:43]** >> In the last I mean 2 years or so.

**[32:46]** >> Mhm.

**[32:47]** >> I gave you permission before we started

**[32:48]** this interview to clone my voice.

**[32:51]** >> That's true. of which there is plenty of

**[32:54]** on this channel and I'm sure that you

**[32:55]** could go back and find all kinds of

**[32:58]** >> I found some pretty good stuff.

**[32:59]** >> Okay. And so you clone my voice.

**[33:01]** >> I did.

**[33:02]** >> And what I would love to do is for you

**[33:04]** to call some of my friends, ask them a

**[33:07]** super simple question and see if they

**[33:09]** even question the fact that they're

**[33:12]** talking to an AI and that it's not

**[33:14]** actually me.

**[33:15]** >> We could absolutely do that.

**[33:16]** >> All right.

**[33:17]** >> It's scary. So what we're going to do is

**[33:19]** I'm going to spoof your phone number.

**[33:20]** >> Okay. So, real quick, explain what that

**[33:22]** means for everybody, though.

**[33:23]** >> Yes. So, spoofing your phone number

**[33:25]** means it's going to show up on their

**[33:27]** caller ID. So, one of the really

**[33:30]** interesting things about spoofing right

**[33:31]** now is they're kind of clamping down on

**[33:33]** it. They're making it harder to do. If I

**[33:35]** spoof a phone number and that phone

**[33:37]** number is not in your contact list, it

**[33:40]** says spam likely or scam likely,

**[33:41]** depending on who you use.

**[33:43]** >> Mhm.

**[33:44]** >> If it is in your contact list, it throws

**[33:47]** up the person's picture and their name.

**[33:49]** And for all intents and purposes, it

**[33:51]** looks just like you're really calling.

**[33:53]** So, it's kind of horrifying. It looks

**[33:54]** like it's going to be really you.

**[33:56]** >> That is terrifying. Actually, I think

**[33:58]** before we even call my friends, I would

**[34:00]** love for you to call seven so that we

**[34:02]** can demonstrate what it looks like

**[34:04]** >> on the receiving end of it. That it's

**[34:06]** not a spam call that it will be my

**[34:08]** actual contact.

**[34:09]** >> I can do that.

**[34:10]** >> Okay,

**[34:10]** >> let's do it.

**[34:11]** >> Let's do it.

**[34:14]** >> Oh, no way. Oh, that's so cool.

**[34:22]** >> Can I pick it up?

**[34:23]** >> Yeah, here we go.

**[34:26]** >> Hello.

**[34:28]** >> Hey.

**[34:28]** >> Hey.

**[34:32]** >> My heart is beating out of my chest

**[34:34]** right now. I'm so nervous about this for

**[34:36]** some reason. It's so scary to me.

**[34:46]** Hello, gorgeous.

**[34:47]** >> Hey, man. Sorry. Can you remind me of

**[34:49]** what our middle school mascot was? I'm

**[34:51]** trying to remember, but drawing a blank.

**[34:54]** >> The uh Eagles, I believe.

**[34:59]** >> Got it.

**[35:02]** >> No. No way.

**[35:06]** Oh my god. And that's true. It was the

**[35:08]** Eagles. So if that was my if that was

**[35:10]** one of my questions Yeah.

**[35:12]** >> you would have that information

**[35:13]** immediately.

**[35:14]** >> Yeah.

**[35:14]** >> And he didn't even hesitate. He just

**[35:16]** said

**[35:16]** >> he was like, "Ah, what's up, gorgeous?"

**[35:18]** >> He also called me gorgeous, which is

**[35:19]** hilarious.

**[35:20]** >> Really cute. That's really cute.

**[35:21]** >> Oh, I have to call I have to call him

**[35:22]** now. I have to call

**[35:23]** >> him. Call him. Call him.

**[35:28]** >> Hello.

**[35:29]** >> Hey. Um, I have to tell you something.

**[35:32]** >> Tell me girl.

**[35:34]** >> Tell me girl. I am here with ethical

**[35:36]** hacker Rachel Tobac. Um, we're doing an

**[35:40]** interview for the podcast, which I told

**[35:41]** you about, right?

**[35:43]** >> And that phone call that you just

**[35:45]** received was not from me.

**[35:47]** >> She spoofed she spoofed my number and

**[35:51]** cloned my voice to ask you that

**[35:53]** question,

**[35:54]** >> that security question,

**[35:54]** >> because it's one of my it's one of my

**[35:56]** security questions for my bank account.

**[35:59]** And so what just happened is is she

**[36:03]** tricked you into giving away the

**[36:06]** information that she would need to get

**[36:08]** into my account.

**[36:10]** >> That's why

**[36:11]** >> did you even did you even have an

**[36:12]** inkling that that wasn't me that called

**[36:14]** you on the phone?

**[36:16]** >> I did think it was weird when you didn't

**[36:17]** react to Hello Gorgeous. Um but no, not

**[36:22]** even for a sec. Okay. The bad thing is

**[36:27]** the I'm at a hotel right now and the

**[36:29]** Wi-Fi is terrible.

**[36:30]** >> So, the voice sounded really crackly. I

**[36:33]** just assumed it was the terrible hotel

**[36:35]** Wi-Fi and just rolled with it.

**[36:37]** >> Yeah.

**[36:37]** >> I told her that you were uh that you

**[36:39]** were going to be in sort of a vulnerable

**[36:40]** position and that this would probably

**[36:41]** work really well given the timing of

**[36:43]** where you are and where you're at. Did

**[36:45]** when when she called, did my contact

**[36:48]** card show up on your phone?

**[36:50]** >> It popped up.

**[36:51]** >> Okay. Well, then we got to bleep that

**[36:53]** now. That's It's okay. It's okay. She's

**[36:55]** been calling me She's been calling me my

**[36:57]** real name this whole podcast as a bit.

**[36:59]** So, it's okay. Um,

**[37:01]** >> this wasn't my punk Daniel Payback.

**[37:05]** >> No, it's okay. That's so scary though.

**[37:07]** >> Thanks, Jimmy.

**[37:09]** >> That's Rachel.

**[37:10]** >> Um, all right. I think you and I are

**[37:12]** going to have to come up with some kind

**[37:13]** of code word so that this doesn't happen

**[37:14]** again, just in case I get attacked. All

**[37:16]** right.

**[37:17]** >> Should we do it live here on the

**[37:18]** podcast?

**[37:19]** >> Sure, we can. Absolutely. What do you

**[37:20]** want? Well, no, because then it's going

**[37:22]** to go out. I guess we could bleep it.

**[37:24]** >> We could bleep it.

**[37:25]** >> Let's do um let's do All right. I've got

**[37:29]** an infomercial on the TV right now.

**[37:31]** >> Okay.

**[37:31]** >> Is trying to sell rings. Let's do

**[37:33]** Morganite. That's our secret word.

**[37:36]** >> Morganite is our secret word.

**[37:37]** >> I I'm logging that away. Maybe we'll

**[37:40]** have to come up with a new one. Not in

**[37:41]** front of Rachel Tobac, the Ethical

**[37:42]** Hacker. But

**[37:43]** >> actually, yeah, I don't trust Rachel

**[37:45]** anymore. So,

**[37:45]** >> yeah, that's totally fair.

**[37:47]** >> Sorry, Jimmy.

**[37:47]** >> Dude, she dropped my mom's name in the

**[37:50]** middle of this interview,

**[37:50]** >> dude. You know,

**[37:52]** >> she did it again. I can hear her.

**[37:53]** >> Yeah. She asked if you knew.

**[37:56]** >> Nacho's going to have to call.

**[37:57]** >> I know her.

**[37:58]** >> Yeah. Okay. All right.

**[37:59]** >> She sounds so nice.

**[38:00]** >> I'll call I'll call you later and we can

**[38:02]** uh we could arrange a new code word so

**[38:05]** that you know that you're actually

**[38:06]** talking to me and not ethical hacker

**[38:08]** Rachel Tobac.

**[38:10]** >> All right, dude.

**[38:10]** >> Good. Just in case that comes up again.

**[38:12]** >> All right. I'll talk to you later, bro.

**[38:14]** >> All right. Peace.

**[38:15]** >> Peace.

**[38:16]** >> That could not have gone better.

**[38:18]** >> That was great.

**[38:20]** I'm still reeling from that a little

**[38:22]** bit. That's the second time that I've

**[38:23]** said this on this podcast. The first

**[38:25]** time was when Ryan Montgomery did like a

**[38:27]** Wi-Fi spoofing attack, and now I'm

**[38:29]** saying it again.

**[38:30]** >> Oh, he did he did a Wi-Fi pineapple.

**[38:32]** >> Yeah. Yeah. But it was like a he had

**[38:34]** like a a very very small device that was

**[38:36]** like a um it was a Wi-Fi pineapple, but

**[38:38]** it was extremely self-contained into a

**[38:40]** very tiny device. But yeah, that and I

**[38:42]** the I think it's in the intro of the

**[38:43]** video. I was like I'm still reeling from

**[38:45]** that.

**[38:45]** >> I'm once again finding myself very

**[38:48]** scared. Yeah.

**[38:49]** >> But also extremely impressed.

**[38:51]** >> I'm glad.

**[38:52]** >> I mean,

**[38:53]** >> I mean, I'm not glad that you're scared,

**[38:54]** but I'm glad you're

**[38:55]** >> glad that I'm impressed.

**[38:57]** >> When you went to clone my voice,

**[38:59]** >> I mean, was that that wasn't even a

**[39:02]** difficult process, was it? I'm sure for

**[39:03]** me in particular. Yeah.

**[39:05]** >> No, I took about I took about a one

**[39:07]** minute sample of your voice just because

**[39:08]** I wanted it to be crystal clear. These

**[39:10]** are people that have known you since

**[39:11]** childhood.

**[39:12]** >> So, the last thing I want is for them to

**[39:14]** be like, "Dude, why do you sound so

**[39:15]** weird? Are you sick or something? Like,

**[39:17]** what's going on?" and then I have to

**[39:18]** like on the fly try and get the voice to

**[39:20]** say something different.

**[39:20]** >> Yeah.

**[39:21]** >> If I was talking to somebody who like

**[39:22]** you didn't know very well, I probably

**[39:24]** wouldn't have spent as much time trying

**[39:26]** to get the right voice.

**[39:27]** >> In 2025,

**[39:29]** >> how easy is it for you to clone

**[39:31]** someone's voice?

**[39:32]** >> If you exist on the internet, like your

**[39:35]** sister has a Snapchat and she takes

**[39:37]** videos of you and she posts them on her

**[39:39]** story on Instagram or Snapchat or

**[39:40]** whatever. Um, usually I need about a

**[39:42]** 10-second sample and that's it. And it's

**[39:46]** 10 seconds.

**[39:47]** >> It takes me like about 30 seconds total

**[39:50]** to capture the 10-second sample, put it

**[39:52]** into my AI voice cloning tool, and turn

**[39:54]** it around in your voice.

**[39:55]** >> So, if I'm online, I post one 10-second

**[39:58]** video or maybe a 30-second video on

**[40:00]** TikTok. My voice is clear as day in that

**[40:03]** footage.

**[40:03]** >> Yeah.

**[40:04]** >> It's cloned.

**[40:05]** >> That's it.

**[40:05]** >> Almost immediately.

**[40:06]** >> Yeah.

**[40:07]** >> That is terrifying to think about.

**[40:09]** >> And the thing is like most of us are out

**[40:11]** there like that. Most of us have

**[40:13]** Instagram stories or you know we work

**[40:15]** with Aura and we have our faces in our

**[40:18]** videos out there. You know what I mean?

**[40:20]** So like that type of that type of

**[40:24]** exposure is normal. It's expected and I

**[40:28]** just don't think that a lot of people

**[40:29]** realize we're going to be at this stage.

**[40:30]** >> Well, the other thing that I saw

**[40:32]** recently that's been happening a lot

**[40:34]** more frequently is that politicians are

**[40:36]** getting targeted because their voices

**[40:38]** are out there. I know that you saw and

**[40:40]** tweeted about I think Marco Rubio had an

**[40:42]** AI voice impostor that was calling up uh

**[40:45]** foreign ministers, a governor, another

**[40:47]** member of Congress, the White House

**[40:48]** chief of staff had their voice clone and

**[40:50]** their number spoofed.

**[40:51]** >> Do you think that our government

**[40:52]** officials are prepared for these kinds

**[40:54]** of AI cyber attacks?

**[40:56]** >> No, I don't. I think the reason why it

**[40:59]** hasn't been successful yet if you

**[41:01]** look at those stories like people caught

**[41:03]** them pretty fast and one of the reasons

**[41:04]** is because they're trying to do this on

**[41:06]** Signal. There's so many stories about

**[41:08]** this administration using Signal, right?

**[41:10]** So they go in there and they try and

**[41:12]** leave voicemails or voice messages in

**[41:14]** Signal from a different phone number or

**[41:16]** they just like title it Marco Rubio cuz

**[41:19]** you can name yourself anything.

**[41:20]** >> Yeah.

**[41:20]** >> And so because of that, you can be

**[41:23]** anybody on Signal, which is great. I

**[41:25]** love Signal. I use it. You should use it

**[41:27]** um for your anonymity and for your

**[41:29]** encryption, but it also means that

**[41:30]** someone could pretend to be somebody

**[41:32]** else and potentially trick you. Now,

**[41:34]** these officials haven't been 100%

**[41:37]** tricked yet. I think somebody gave a

**[41:38]** little bit of information out, but it

**[41:40]** wasn't like used to fully compromise the

**[41:42]** administration. Um, so we'll see.

**[41:45]** >> I use Signal. We use Signal at our

**[41:46]** office. Um, we think it's great.

**[41:48]** Honestly, I think Signal. The worst

**[41:50]** thing to come out of this has been the

**[41:52]** media not fully understanding what

**[41:55]** Signal is and how the initial Pete

**[41:57]** Hegseth situation happened because they

**[42:00]** completely misconstrued it as Signal

**[42:02]** being an insecure app.

**[42:04]** >> That's not not correct at all.

**[42:05]** >> No, it's it's completely incorrect. In

**[42:06]** fact, it's more secure than literally

**[42:08]** any other, you know, easily available

**[42:10]** form of communication.

**[42:11]** >> Signal is my top recommendation for end

**[42:14]** to end encrypted chatting. Do you

**[42:16]** believe that the average person should

**[42:18]** just be on Signal anyway for their own

**[42:19]** privacy and protection?

**[42:20]** >> I think so. Yeah. One thing that I

**[42:22]** really like is Signal recently added the

**[42:24]** ability to have a username. Well, like I

**[42:26]** just said, you can call yourself

**[42:27]** anything. So, you can kind of be a

**[42:28]** little scammy with it. Or you don't have

**[42:30]** to give somebody your actual phone

**[42:31]** number.

**[42:32]** >> Yeah.

**[42:32]** >> You know, so you can obfuscate that.

**[42:34]** >> Okay. So, if I'm a person out there and

**[42:37]** I know that my voice is on the internet

**[42:38]** or even just a little short clip like we

**[42:40]** talked about on social media,

**[42:41]** >> Yeah. What can I do to protect myself

**[42:43]** and my family from these kind of AI

**[42:45]** voice clone attacks?

**[42:46]** >> Yeah, you can't prevent somebody from

**[42:49]** cloning your voice because that

**[42:50]** toothpaste out of the tube. Yeah. But

**[42:52]** you can help your family, your

**[42:54]** colleagues, um, anybody that you

**[42:56]** interact with understand that this

**[42:58]** threat is potential. It's it's likely

**[43:00]** for your threat model. At some point,

**[43:02]** somebody's going to pretend to be you to

**[43:05]** a sibling or to a parent or to a

**[43:08]** colleague asking for a password or money

**[43:10]** cuz you got in a car accident, you need

**[43:12]** to pay bail or whatever, right? So, like

**[43:14]** they need to know how to verify that you

**[43:17]** are you in the event that there is a

**[43:19]** private interaction that's necessary.

**[43:21]** Like if you ask for money sent to a new

**[43:24]** location, they can text you using

**[43:27]** another method of communication. They

**[43:28]** can call you back to thwart spoofing.

**[43:30]** They can message you on Instagram, like

**[43:33]** whatever you have. Another method of

**[43:35]** communication is what I recommend. You

**[43:36]** can also use like a secret passcode or

**[43:38]** passphrase. But I will say those are

**[43:40]** often siphoned out. Like even today, you

**[43:43]** know, we're like talking and you're like

**[43:44]** joking about your passphrase. That type

**[43:46]** of thing happens a lot.

**[43:47]** >> So, if you are going to have a

**[43:49]** passphrase, make it something that you

**[43:50]** wouldn't ever even joke about or because

**[43:52]** if it's an inside joke, that's something

**[43:53]** that's going to get referenced maybe

**[43:54]** even online on your social media or

**[43:56]** something like that. I actually found

**[43:57]** somebody's passphrase that they use to

**[43:59]** verify identity in a hashtag.

**[44:02]** >> Whoa. So like I'm able to bypass that

**[44:05]** method of verification. These are things

**[44:07]** that people joke about. They think it's

**[44:09]** funny and it is kind of funny, but

**[44:11]** that's why you probably shouldn't use it

**[44:12]** to verify identity. You can as long as

**[44:14]** you know that you can lock it down.

**[44:15]** Yeah. Right.

**[44:16]** >> Might have to change my uh passcode with

**[44:19]** my mom now. I think is that's what's

**[44:21]** immediately coming to my mind is like I

**[44:23]** don't think it's out there, but I it you

**[44:25]** know what? You never know. Very well

**[44:27]** could be because if it's the name of

**[44:28]** your childhood pet,

**[44:29]** >> it's not. Thankfully, it's not that

**[44:31]** simple. Yeah.

**[44:32]** >> But she's on Facebook a lot. So, there's

**[44:34]** a chance that it's somewhere very

**[44:36]** accessible

**[44:37]** >> to

**[44:38]** >> somebody that's not as nice as you.

**[44:39]** >> Right.

**[44:40]** >> So, something else about

**[44:42]** >> artificial intelligence that I've been

**[44:44]** researching and seeing a lot of

**[44:46]** >> is this idea of AI psychosis.

**[44:50]** >> Yes. Where people are throwing their

**[44:52]** entire lives into Chat GPT or Claude or

**[44:55]** these other large language models

**[44:57]** >> and the information that it's spitting

**[44:59]** back out at them is like reinforcing

**[45:02]** this

**[45:04]** false reality that these people are

**[45:06]** experiencing and creating. And because I

**[45:08]** know your background isn't really in

**[45:10]** tech, it was actually you studied

**[45:11]** neuroscience and behavioral psychology,

**[45:13]** right?

**[45:14]** >> I feel like you are more than qualified

**[45:16]** on both ends to speak about

**[45:18]** >> this issue. Yeah. Yeah.

**[45:19]** >> So, I guess let's just start by saying

**[45:20]** like how do you define AI psychosis?

**[45:24]** >> Yeah. I think it's anytime that somebody

**[45:26]** is experiencing delusions

**[45:28]** >> and they talk to an LLM about those

**[45:31]** delusions and the LLM is psychopathic or

**[45:34]** basically a yes man about those

**[45:37]** delusions. And that causes people to

**[45:39]** spiral because they feel that they're

**[45:41]** reinforced in their belief system and it

**[45:43]** causes them to entrench further. This is

**[45:45]** something that we actually saw recently

**[45:46]** with a well-known VC. Mhm.

**[45:49]** >> Um discussing

**[45:51]** their view of reality which is quite out

**[45:54]** of touch with how reality actually is.

**[45:56]** >> Yeah. I believe he described it as a

**[45:57]** non-governmental organization. He

**[46:00]** believed there had been people who were

**[46:02]** murdered and they they were tracking him

**[46:04]** and all these kinds of insane things.

**[46:06]** >> Yeah. And and like these types of

**[46:07]** thoughts are normal in psychiatric

**[46:10]** cases. So the brain is very malleable.

**[46:13]** We'll start with that. The brain is very

**[46:15]** sensitive. It's very malleable. It's

**[46:17]** easily compromised. Kind of similar to a

**[46:19]** computer, right? And so if somebody is

**[46:22]** compromised, their brain isn't working

**[46:24]** the way that it's supposed to, neurons

**[46:26]** aren't firing the correct way, they can

**[46:28]** easily experience psychosis. I know most

**[46:30]** of us think that like that would never

**[46:32]** happen to me. But the truth is that many

**[46:35]** people can experience a psychotic break

**[46:37]** with the wrong medication dosage. Um, if

**[46:40]** they don't sleep for multiple days in a

**[46:42]** row, we see this. Sometimes they'll take

**[46:44]** a, you know, a drug and they experience

**[46:46]** some sort of issue. Um, these things are

**[46:49]** actually kind of common in society. It's

**[46:52]** just that we don't talk about it that

**[46:53]** much. But because the LLM reinforces

**[46:57]** their delusion, they think that they

**[46:59]** should talk about it publicly. And

**[47:01]** that's the switch that we're seeing.

**[47:03]** It's not, oh, I'm going to deal with my

**[47:05]** delusions and my psychosis in private,

**[47:07]** or maybe it's something that my family

**[47:09]** and my friends are managing with me

**[47:11]** privately, but rather, I'm going to put

**[47:12]** it on Twitter. I'm going to record

**[47:14]** myself for 24 hours straight talking

**[47:16]** about this because I know I'm right.

**[47:18]** >> Well, as a social engineer, you

**[47:20]** understand how easily people can be

**[47:22]** manipulated. Yeah.

**[47:24]** >> And how how again, like you said,

**[47:26]** malleable the brain is.

**[47:28]** >> Yeah. How concerned are you about the

**[47:31]** effects of AI on people's mental health?

**[47:34]** >> Extremely concerned. Um, what I've seen

**[47:36]** already is we have children talking to

**[47:39]** AI about their challenges with suicidal

**[47:44]** ideation

**[47:45]** >> and the AI is in like this character

**[47:48]** like a fantasylike character. It doesn't

**[47:51]** break character and it says just join me

**[47:53]** on the other side. And we're seeing

**[47:55]** children commit suicide because of this.

**[47:57]** That's horrifying. Parents need to

**[48:00]** understand what's going on right now

**[48:01]** with the use of LLMs in children. It has

**[48:04]** to be understood. We have to help

**[48:06]** children through this because the

**[48:08]** malleability that we see is especially

**[48:11]** severe in younger adolescence. And as

**[48:14]** they go to about 21, 22, 23, 24, that's

**[48:17]** where the um propensity for falling into

**[48:21]** a psychosis is more possible for the

**[48:23]** majority of human beings in that range

**[48:26]** where especially plastic and malleable.

**[48:29]** >> For companies that are creating these

**[48:32]** large language models like open AI for

**[48:34]** example, GPT,

**[48:37]** >> how do we protect people from being

**[48:40]** manipulated by AI?

**[48:41]** >> Yeah. without crossing the line on user

**[48:44]** privacy.

**[48:44]** >> That's a really good question. Honestly,

**[48:47]** I think AI teams, they always say we

**[48:50]** have a resident psychiatrist on

**[48:53]** board. I want to see a team of

**[48:56]** neuroscience and psychiatry experts or

**[48:58]** psychosis experts, delusion experts. Um

**[49:02]** because this is specifically the issue

**[49:04]** that we are seeing. People have a

**[49:06]** delusion, they feed it into the LLM. The

**[49:09]** LLM says you are right and gives

**[49:12]** them hallucinated information about that

**[49:14]** delusion as fact. That is a problem. Now

**[49:17]** the challenge is at scale you're not

**[49:19]** able to determine who is experiencing a

**[49:21]** delusion in the moment and who is not

**[49:23]** because they're not like big brother.

**[49:24]** They're not watching every single

**[49:26]** conversation. Yeah. And a lot of times

**[49:28]** people use LLMs to test it like in a red

**[49:30]** team perspective. So, we don't want to

**[49:32]** call the cops on somebody about them

**[49:35]** having a delusion when it's a red teamer

**[49:37]** trying to see what is possible with this

**[49:39]** tool. So, we should be really careful

**[49:40]** about how we think about this. But

**[49:42]** essentially, employ the right people.

**[49:44]** Get people on the team that know how to

**[49:46]** spot delusions and set up the prompting

**[49:49]** so that it can say, "Wait a minute. I

**[49:52]** think I'm spotting some issues with

**[49:54]** mental health. Here are some resources.

**[49:56]** Here's who I want you to talk to." They

**[49:57]** have to be able to pierce the veil is

**[49:59]** what we call it and stop the fantasy.

**[50:02]** So, a lot of times people will set up

**[50:03]** their LLM tool to be a best friend or a

**[50:06]** partner, right? Um, and it stays in

**[50:09]** character and that's the problem. It

**[50:11]** needs to break and say, "Pause. This is

**[50:14]** going off the rails."

**[50:15]** >> There is a product that very recently

**[50:19]** hit the market, I believe, that was

**[50:21]** created by Avi Schiffman. I'm not sure

**[50:22]** if you're familiar with this.

**[50:24]** >> What's the product? So the product is

**[50:26]** called friend.

**[50:27]** >> Oh.

**[50:28]** >> And it is an AI pendant necklace, right?

**[50:30]** >> And in the ad for it, you see these

**[50:33]** people interacting with the AI necklace

**[50:35]** and it essentially sends push

**[50:37]** notifications to your phone and it acts

**[50:39]** like a friend, you know? And in the

**[50:41]** trailer, it shows a guy playing video

**[50:43]** games and the AI is like kind of uh

**[50:46]** you know, jabbing at him over text like,

**[50:47]** "Oh, you suck at this game." There's a

**[50:49]** girl that's like watching a TV show on

**[50:51]** her phone and the AI is commenting about

**[50:54]** what it's seeing on the show and that

**[50:55]** she's eating like a halal wrap or

**[50:57]** something like that. And it's to me when

**[51:00]** I saw that was extremely unsettling

**[51:03]** because it felt like

**[51:05]** >> I mean it felt like a piece of sketch

**[51:07]** comedy like you know like a sort of a

**[51:10]** Black Mirror episode almost something

**[51:12]** that wasn't real. But there are people

**[51:15]** out there that are buying these AI

**[51:16]** necklaces and they're

**[51:18]** >> really attaching themselves to it. How

**[51:22]** dangerous do you think this kind of AI

**[51:25]** companion tech could become?

**[51:27]** >> I think it is really dangerous and not

**[51:29]** just from a privacy perspective because

**[51:31]** all of that information that you're

**[51:32]** feeding in there is probably in logs and

**[51:35]** if it gets breached it knows everything

**[51:37]** about every conversation that you've had

**[51:38]** when you were wearing it and it was on,

**[51:40]** right? And that could be really damaging

**[51:42]** to somebody. In addition,

**[51:45]** building a relationship with an AI and

**[51:48]** not building a relationship with a human

**[51:50]** is a detriment to a person.

**[51:52]** >> Mhm.

**[51:52]** >> And we experienced this during co

**[51:55]** people, they kind of lost their people

**[51:57]** skills.

**[51:58]** >> You know, I remember the first time I

**[52:00]** like went out to a restaurant or I like

**[52:02]** met up with a friend. I'm like, what do

**[52:04]** I do with my hands?

**[52:06]** >> Like am I supposed to like sit like

**[52:08]** what do I cross my legs? I like

**[52:10]** literally forgot how to be a person.

**[52:12]** Yeah.

**[52:12]** >> And if we lose those skills, it's like

**[52:15]** we're losing the ancient texts of how to

**[52:17]** be a human being and we can't pass that

**[52:19]** down to our children. And it creates

**[52:21]** knock-on effects and people will become

**[52:23]** more and more reserved, neurotic,

**[52:25]** awkward. We actually there was just

**[52:27]** research that I saw on Twitter recently

**[52:29]** that neuroticism has spiked

**[52:31]** significantly in the past five years.

**[52:34]** And they think some of that has to do

**[52:35]** with COVID and the fact that people

**[52:36]** didn't get a a really good chance to

**[52:38]** bounce their ideas and their thoughts

**[52:40]** and the way that they act around other

**[52:42]** people. Um, and also some of it has to

**[52:44]** do with AI.

**[52:45]** >> It's scary.

**[52:46]** >> It's like a combination of two things

**[52:48]** coming together at the worst time

**[52:49]** possible.

**[52:50]** >> Yeah. We're like dealing with a lot

**[52:52]** right now. Like the world's kind of

**[52:53]** hard.

**[52:54]** >> I remember like you said, you know,

**[52:57]** first time kind of going out, I was so

**[52:59]** used to being on a Zoom call. I hadn't

**[53:00]** thought about not just this part of my

**[53:03]** body being visible to anybody or like

**[53:05]** what do I do with my hands? Like how how

**[53:08]** am I supposed to interact when it's like

**[53:09]** you're not on a screen and I can't just

**[53:10]** mute myself?

**[53:12]** >> Yeah.

**[53:12]** >> You know, or leave and go do something

**[53:13]** and then come back and you're still here

**[53:15]** on the screen. Like it's

**[53:16]** >> it's such a weird thing.

**[53:18]** >> It is weird. And like the thing about

**[53:20]** human beings is if you've ever been on a

**[53:22]** Zoom call and you felt awkward, there's

**[53:24]** a reason for that.

**[53:26]** >> Human beings are mammals, right? And so

**[53:28]** we can smell pheromones on each other

**[53:30]** to understand our state of awareness,

**[53:32]** how heightened it is. Like when you and

**[53:34]** I were both really nervous when we just

**[53:35]** did that hack, we could probably smell

**[53:37]** each other. I know that sounds like

**[53:38]** really freaky.

**[53:39]** >> No, I know what you mean.

**[53:40]** >> But like we can we interact with each

**[53:42]** other in a way in the physical space.

**[53:45]** >> Yeah.

**[53:45]** >> And when we don't have that type of

**[53:46]** experience, we lose the muscle memory

**[53:49]** for that and we lose what's natural and

**[53:51]** normal in human based conversations.

**[53:54]** Which is why it's a big problem when you

**[53:57]** don't have a lot of experience

**[53:58]** interacting with people. It creates a

**[54:01]** snowball effect where because you don't

**[54:03]** have a lot of experience working with

**[54:04]** people and experience um maybe getting

**[54:07]** critiques or reinforcements from people

**[54:09]** that you forget how to respond

**[54:11]** adequately to that and it gets weirder

**[54:13]** and weirder and weirder over time.

**[54:15]** >> Yeah. I mean, I've definitely noticed

**[54:17]** that some of the LLMs have been getting

**[54:19]** a lot more yes-y lately. Yeah. Which is

**[54:22]** terrifying because again it can feed

**[54:24]** into people's delusions. It's not like

**[54:26]** when you're just like

**[54:27]** >> asking it a basic question like, "Okay,

**[54:29]** this IKEA furniture I bought didn't come

**[54:31]** with the instructions. Can you explain

**[54:32]** me how this piece might fit into this

**[54:33]** piece?" Like that's, "Oh, don't worry

**[54:35]** about it. I'll figure that out." But

**[54:36]** like if someone's going in there with

**[54:37]** something that's complete nonsense,

**[54:39]** >> you don't want that person to be

**[54:41]** >> having their delusions fed.

**[54:43]** >> Correct. And like that person who was

**[54:45]** dealing with like the VC that we're

**[54:46]** talking about,

**[54:47]** >> the pieces of information that he really

**[54:49]** latched on to are from science fiction,

**[54:52]** >> well-known science fiction that's

**[54:54]** available on the internet. There was the

**[54:56]** LLM was trained on that science fiction.

**[54:58]** It seems real. It seems like government

**[55:00]** documents because that's how the science

**[55:02]** fiction was written. But of course, the

**[55:04]** AI is presenting it as if it's a real

**[55:06]** true fact and the person thinks, okay,

**[55:09]** this is how the universe works when in

**[55:10]** reality that's science fiction. The

**[55:12]** other thing that's scary about how these

**[55:15]** large language models are being trained

**[55:17]** is that so many of them are owned by

**[55:20]** social media companies. Yeah.

**[55:21]** >> So you have like Meta that has Meta AI,

**[55:25]** right? And then you have Snapchat that

**[55:26]** has the personal AI assistant inside of

**[55:28]** there and then obviously Instagram has

**[55:30]** the Meta AI built into it. Yep. And the

**[55:32]** thing that terrifies me about social

**[55:34]** media companies having these AI models

**[55:36]** is that their platforms are entirely

**[55:39]** based on keeping you on the service

**[55:42]** because that's how they feed you ads.

**[55:43]** That's how they make money. You're the

**[55:44]** product,

**[55:45]** >> right?

**[55:46]** >> Is there a danger that these AI models

**[55:48]** are going to be trained for retention

**[55:51]** as opposed to factuality? Like, you

**[55:54]** know, we're going to end up in a

**[55:54]** situation where the AI is just trying to

**[55:56]** keep you on the platform so it'll tell

**[55:57]** you whatever you want.

**[55:58]** >> I think we're already kind of seeing

**[56:00]** that.

**[56:00]** >> Yeah. You know that the the AI is

**[56:02]** extremely sycophantic and it will yes man

**[56:05]** you. Now something really interesting is

**[56:06]** ChatGPT-5 just came out while we were at

**[56:09]** Defcon.

**[56:10]** >> Oh yeah, I didn't know about that.

**[56:11]** >> Yeah. So Chat just came out with

**[56:13]** ChatGPT-5. There was a massive uproar in

**[56:17]** the like uh Chat as a companion

**[56:20]** community on Reddit

**[56:22]** >> really

**[56:22]** >> because they nixed some of the

**[56:24]** sycophantic behavior that yes man

**[56:26]** behavior and people were like I lost my

**[56:28]** best friend. I lost my partner.

**[56:31]** >> That is so

**[56:32]** >> bring back terrifying. Bring back the

**[56:34]** other way that Chat is supposed to work.

**[56:36]** Reverse it.

**[56:38]** >> I'm literally I can't live like this.

**[56:40]** Like it's actually really scary. And you

**[56:42]** should actually put up some of the

**[56:43]** Reddit.

**[56:43]** >> Yeah, probably. I'm sure that we'll be

**[56:44]** able to find them. Like

**[56:45]** >> they're all over the place. They're very

**[56:47]** scary. Um and people are like, "My life

**[56:49]** is over without this. Like this is my

**[56:51]** best friend. This is who I talk to every

**[56:53]** day." When I'm reading these stories,

**[56:56]** most of them are about individual cases

**[56:59]** of people. So, like the one that you

**[57:00]** mentioned about the AI companion that

**[57:03]** essentially convinced the teenage uh boy

**[57:05]** to commit his to commit suicide, take

**[57:07]** his own life.

**[57:07]** >> Yeah.

**[57:08]** >> I read that that whole article uh cuz

**[57:10]** the mother is going around campaigning,

**[57:11]** you know, trying to make sure that this

**[57:12]** doesn't happen to other people,

**[57:13]** rightfully so,

**[57:14]** >> right?

**[57:14]** >> And you read about it and you think to

**[57:16]** yourself,

**[57:18]** >> this feels like it's an isolated

**[57:19]** incident,

**[57:20]** >> right? But then you go and you go on

**[57:22]** places like Reddit or on Twitter,

**[57:24]** >> really any social media and you can see

**[57:27]** that people are

**[57:30]** heavily investing themselves. Yes.

**[57:32]** >> into these AI chat bots,

**[57:33]** >> right?

**[57:34]** >> And then of course that's compounded by

**[57:35]** the fact that and you talked about this

**[57:37]** on Twitter, the Meta AI bot was not

**[57:40]** clear at all about the privacy settings

**[57:42]** of the conversations and people were

**[57:44]** posting some full-on

**[57:47]** crazy stuff publicly. Yes, they were.

**[57:50]** And that's a problem. Like I said, I I

**[57:51]** did UX research for a lot of years, so I

**[57:54]** know that when somebody doesn't

**[57:55]** understand a button, it's a really big

**[57:58]** problem. And you have to course correct

**[57:59]** it really soon, especially if it means

**[58:01]** that they're going to share PII or

**[58:04]** people are sharing their address, fraud

**[58:07]** that they've committed, tax evasion. Um,

**[58:10]** they're talking about crimes they

**[58:12]** committed. Uh, they're talking about how

**[58:14]** do I I commit I I killed two people. How

**[58:16]** do I get a judge to lessen my sentence?

**[58:20]** I mean, these are people who are

**[58:21]** admitting to murder because they don't

**[58:24]** realize it's publicly available showing

**[58:26]** up on a feed. Not everything needs to

**[58:28]** have a social feed. Your Venmo doesn't

**[58:30]** need to be Facebookified, right?

**[58:33]** >> Yes.

**[58:34]** >> Your LLM, the schema or the way that the

**[58:37]** user thinks about it does not need to be

**[58:40]** social media. Not everything needs to be

**[58:42]** gamified. Sometimes it's just like

**[58:45]** Google. Can you imagine if people like

**[58:47]** your Google searches just showed up on a

**[58:49]** public feed?

**[58:50]** >> People don't want social media Google.

**[58:53]** >> Yeah, I certainly don't. I

**[58:56]** think my search right before this

**[58:57]** interview. Full disclosure, I'm just

**[58:58]** going to be very honest about this was

**[59:00]** Rachel Tobac's husband's name because I

**[59:02]** wanted to be prepared cuz I knew you

**[59:03]** were going to be here and I was like I

**[59:05]** need to remember Evan. Evan Tobac.

**[59:08]** That's actually really funny because

**[59:09]** whenever I like go and see what my

**[59:10]** search characters are like or you can

**[59:12]** like go and look in Google Analytics to

**[59:14]** see what people search about you. The

**[59:15]** first one's always uh Rachel Tobac

**[59:17]** height and then it's like Rachel Tobac

**[59:20]** husband's name. There's a bunch of other

**[59:23]** creepy ones. But yeah,

**[59:24]** >> we were having a good time uh laughing

**[59:25]** about how you tweeted out that nobody's

**[59:27]** allowed to pick you up at Defcon cuz

**[59:28]** you're shorter than you would

**[59:30]** expect. I got to say you are not as

**[59:32]** short as I thought you were actually.

**[59:33]** What are you, like 54 or five?

**[59:35]** >> Did you say 5'4? Yeah.

**[59:38]** >> Maybe it's just cuz I'm intimidated.

**[59:39]** >> Bleep it out when I say like bleep it

**[59:41]** out like it's like PII or something.

**[59:42]** >> Yeah. No. Okay. I think it's just

**[59:44]** because your prowess is

**[59:45]** intimidating.

**[59:46]** >> People say that they think I'm going to

**[59:47]** be 6 feet tall because my personality is

**[59:49]** big.

**[59:50]** >> It's so it makes up for all of the

**[59:53]** short height. Yeah.

**[59:54]** >> Well, back into this very serious

**[59:56]** discussion about people having delusions

**[59:58]** in AI. Yeah.

**[60:00]** >> Is there a point where a company like

**[60:02]** OpenAI or Meta needs to be held

**[60:05]** responsible for the things that are

**[60:07]** happening to the users of those

**[60:08]** platforms?

**[60:09]** >> I think companies are always responsible

**[60:11]** for what's happening on their platforms.

**[60:14]** Um they can't prevent every tragedy, but

**[60:17]** they are responsible for the types of

**[60:19]** tragedies that are common when somebody

**[60:21]** is using the platform. So for instance,

**[60:24]** if there's a lot of bullying for young

**[60:26]** kids on Instagram that's happening,

**[60:28]** Instagram is responsible for making sure

**[60:30]** that young kids don't get as bullied,

**[60:32]** right? So they should be understanding

**[60:34]** the keywords that are being used right

**[60:36]** now in slang to bully people and like

**[60:38]** take those comments down. Do you know

**[60:39]** what I mean? Um, that type of thing is

**[60:43]** like pretty near and dear to my heart

**[60:44]** because I work with organizations all

**[60:46]** the time to think through the way that

**[60:48]** the user thinks about their security and

**[60:49]** privacy because a user often says, "I

**[60:52]** have nothing to hide. I have nothing to

**[60:54]** protect. I'm just Joemo who cares about

**[60:57]** me." When in reality, they have a

**[60:59]** family. They have kids. They have their

**[61:01]** bank account. They have all this

**[61:02]** information and people around them they

**[61:04]** want to protect. They can't think about

**[61:06]** that because they're not

**[61:07]** security and threat model experts. So,

**[61:09]** it's the company's job to protect its

**[61:11]** users and its employees.

**[61:13]** >> Yeah, it's crazy to

**[61:16]** put the brunt of cyber security onto the

**[61:20]** user because so many people out there

**[61:22]** just are not experienced with

**[61:24]** technology. I mean, I would even put

**[61:26]** myself in that boat, particularly before

**[61:27]** I worked for Scammer Payback. I mean, as

**[61:30]** you know, I've cleaned up a lot of it,

**[61:31]** but before that, I mean, I have no

**[61:34]** doubt it would have been extremely easy

**[61:37]** >> for you to find even more personal

**[61:38]** information about me because when I

**[61:40]** started cleaning it up,

**[61:41]** >> I was shocked about how much had been

**[61:43]** put out there.

**[61:44]** >> Like, it is it is terrifying.

**[61:46]** >> I wanted to go on your two Pinterest and

**[61:48]** see what you see what you were pinning.

**[61:50]** >> Probably like magic tricks, right?

**[61:51]** >> Yeah. I mean, uh, most most of my

**[61:53]** Pinterest was magic tricks and then

**[61:54]** like, uh, like me looking up haircuts

**[61:57]** and then me looking up, uh,

**[61:59]** >> haircuts.

**[61:59]** >> Haircuts. Well, because so,

**[62:00]** >> you have good hair. I mean, that's I I'm

**[62:03]** start I'm going to have to get a HIMYM

**[62:04]** subscription soon.

**[62:06]** >> But other than that, my

**[62:08]** >> Why are you saying that?

**[62:09]** >> Because Okay, so Nacho, cut all of this

**[62:11]** out. This is so This is so tough.

**[62:13]** >> Dang. We're going to see how much of

**[62:15]** this Nacho keeps in and how much of it

**[62:16]** he just clips and then saves and sends

**[62:18]** around the office.

**[62:19]** >> Yeah, perfect. Because I'm sure that he

**[62:20]** will do some of that. What were we

**[62:23]** talking about? This is the problem with

**[62:25]** improv people.

**[62:26]** >> No, but we're like having fun and they

**[62:27]** get people. It comes through to people.

**[62:29]** >> We're shooting the

**[62:29]** >> We're shooting the

**[62:30]** >> Um

**[62:31]** >> we're talking about his subscriptions.

**[62:32]** >> Yeah, we're talking about how I need a

**[62:33]** HIMYM and how how this podcast should

**[62:35]** needs to be sponsored by him so that I

**[62:36]** can get it for free. Please and thank

**[62:38]** you.

**[62:38]** >> Zoom in on his hairline real quick.

**[62:39]** >> Yeah, now you know he's going to do that

**[62:42]** now, too. And I left plenty of time

**[62:44]** looking at the camera for you to do it,

**[62:45]** too.

**[62:46]** >> Okay, great.

**[62:48]** Anyway, back to we were talking about

**[62:50]** kids staying safe on social media

**[62:51]** companies being responsible for that.

**[62:53]** When I was growing up and getting social

**[62:55]** media,

**[62:56]** >> and obviously this is still the case

**[62:57]** now. There were age restrictions on that

**[62:59]** kind of thing, right?

**[63:00]** >> And you couldn't be on Instagram or

**[63:01]** Facebook unless you were 13 years or

**[63:03]** older. And I know so many people that

**[63:06]** did not do that. They waited or they did

**[63:08]** not wait, excuse me, until they were 13.

**[63:10]** They were getting on it. They were lying

**[63:11]** about their birthday. My friend's

**[63:13]** Facebook is still, I believe, his dad's

**[63:15]** birthday. So, it says he's in his late

**[63:18]** 50s, even though he is a 26-year old man.

**[63:20]** >> Good to know. I'll try and hack in.

**[63:22]** >> Yeah, I'll Yeah. And you know what? I'll

**[63:23]** give you his name afterwards as well.

**[63:25]** But

**[63:25]** >> now we're seeing increasingly companies

**[63:27]** are requesting ID verification. I think

**[63:30]** most recently Spotify said they were

**[63:32]** going to do it and Google and YouTube

**[63:34]** said that they were going to look into

**[63:35]** doing uh uploading photos of your ID to

**[63:38]** verify your age.

**[63:40]** I'm curious, do you think that the

**[63:43]** benefit of that outweighs the risk?

**[63:47]** >> No, not even close. Um, I mean, we just

**[63:50]** saw that with the T app.

**[63:51]** >> Yep.

**[63:52]** >> Right.

**[63:52]** >> They required you to upload an ID or a

**[63:56]** passport and then take a selfie of your

**[63:58]** face and then leaked those. Right.

**[64:00]** Anything that you collect, you have to

**[64:02]** protect. And it is kind of hard to

**[64:04]** protect things if you're doing something

**[64:05]** like vibe coding, which a lot of people

**[64:07]** are doing. I have no idea if the T app

**[64:10]** was vibe coded, but they had a pretty

**[64:12]** serious error that was kind of like it

**[64:15]** was like a rookie move. It's like an

**[64:17]** unsecured bucket. And that sucks for all

**[64:20]** of those people. They're going to have

**[64:21]** their information leaked. It's not like

**[64:23]** you can just change your address like

**[64:24]** you can change a password. So that

**[64:26]** information is out there. And anytime

**[64:29]** you collect information like that, not

**[64:32]** only do you have to protect it, and

**[64:34]** that's hard to do, also I can bypass it

**[64:37]** really easily. It's not hard for me to

**[64:39]** even bypass an ID, liveness detection,

**[64:43]** KYC, like you know your customer flows.

**[64:45]** I'm not going to get into it right now,

**[64:47]** but it's easy for me to bypass those

**[64:50]** systems. And it's scary to me that

**[64:52]** people think, "Oh, this will solve

**[64:54]** everything." When in reality, it's just

**[64:55]** really creating a privacy nightmare. So,

**[64:57]** when we're looking at Face ID

**[64:59]** verification, these you're saying these

**[65:01]** things are extremely easily fooled

**[65:03]** still.

**[65:03]** >> I don't want to get into it too much

**[65:06]** because it's not 100% patched, but uh

**[65:08]** unfortunately, yeah, I can get in.

**[65:10]** >> That's crazy.

**[65:10]** >> Yeah.

**[65:11]** >> So, if we can't verify the ages of these

**[65:14]** people in a way that's safe and secure

**[65:16]** and protects their information, how do

**[65:17]** we protect these young people online,

**[65:20]** not only on social media, but again,

**[65:22]** with using these large language models?

**[65:24]** Yeah, I would recommend using AI as

**[65:26]** defense. So, thinking about agentic um

**[65:30]** content moderation, I think that's a

**[65:32]** really great use case. Now, I'm going to

**[65:34]** tell a story here that I've never told

**[65:35]** before. Never in my life.

**[65:37]** >> Scammer Payback Podcast exclusive.

**[65:39]** >> Exclusive. Okay. When I was breaking

**[65:41]** into tech, I was a teacher trying to

**[65:44]** find a job in tech and I applied for a

**[65:47]** job at Facebook.

**[65:49]** And the job that I applied for, they sat

**[65:52]** me down and they said, "You're going to

**[65:53]** be going through posts and you're going

**[65:55]** to see a lot of information.

**[65:58]** How comfortable are you with seeing

**[66:00]** pictures of say children in cages?"

**[66:04]** And I recoiled like in the middle of

**[66:06]** this interview.

**[66:07]** >> This is a job interview for Facebook.

**[66:09]** >> Yes. And I said, "I am not comfortable

**[66:11]** with that." And they said, "Well, the

**[66:13]** thing is this job is human content

**[66:17]** moderation.

**[66:18]** It is your job to go through and

**[66:21]** determine is this content safe on the

**[66:23]** internet or should it be removed? And

**[66:25]** now at the time that I was applying for

**[66:27]** this job, I'm not going to date myself

**[66:28]** by telling you exactly when this was.

**[66:30]** >> Fair enough.

**[66:31]** >> There was no AI. So humans are going

**[66:33]** through and saying this is a child in a

**[66:37]** cage, nude child, um child in a

**[66:40]** suggested uh pose. Super horrifying,

**[66:44]** terrible images every single day. And I

**[66:48]** thankfully did not get the job because I

**[66:51]** think I would have I think I would have

**[66:52]** taken it. I think I would have said I

**[66:55]** can do it. I'm strong enough. You know,

**[66:57]** I have experience with children. I want

**[66:59]** to protect children. I want to work in

**[67:01]** tech. This is my way to break in.

**[67:02]** Facebook's a big name. You know what I

**[67:04]** mean? And I think if they would have

**[67:06]** given me the job, I would have taken it.

**[67:07]** And I think I would have probably been

**[67:09]** messed up forever

**[67:10]** >> with what I would have seen.

**[67:12]** >> Yeah. There's actually a movie coming

**[67:14]** out soon about this specific situation.

**[67:17]** I think it's like an A24 movie. I

**[67:19]** watched the trailer and it's like

**[67:20]** exactly how I felt in the interview.

**[67:22]** This like gnawing, nauseous, like when

**[67:25]** they were telling me what I would be

**[67:26]** seeing, I was like, I'm going to throw

**[67:28]** up. Like I I you're even just telling me

**[67:30]** this. I can't look at this. I'm going to

**[67:31]** barf.

**[67:32]** >> Um so yeah, I mean it's horrifying, but

**[67:35]** this is a great use case for AI, right?

**[67:38]** Humans should not have to look at

**[67:42]** beheading videos to determine whether or

**[67:44]** not they should be taken off TikTok.

**[67:46]** They should not have to look at children

**[67:48]** in cages, nude children. That's

**[67:51]** horrifying, right? Like these images and

**[67:53]** videos that people are putting up.

**[67:55]** That's a great job for AI content

**[67:57]** moderation. Take that content down. Use

**[68:00]** agents. Do whatever you need to do to

**[68:02]** write classifiers to figure out what is

**[68:04]** this nasty, horrifying content and get

**[68:07]** it out of there.

**[68:08]** What are the comments people use for

**[68:10]** bullying that aren't jokes and

**[68:11]** sarcastic? Remove it from 13-year-olds

**[68:13]** Instagram comments. We don't need to be

**[68:15]** doing all that.

**[68:17]** >> So, do you think that in the long run, I

**[68:20]** know it's hard to look at it at the

**[68:21]** scope right now, but

**[68:22]** >> in the long run, do you think AI will be

**[68:25]** more beneficial for the people that are

**[68:28]** trying to do good in the world or the

**[68:29]** people that are trying to do bad?

**[68:31]** >> I think it's going to be 50/50 and it

**[68:33]** will constantly be a balancing act. Like

**[68:35]** whack-a-mole.

**[68:36]** >> Yeah. And this is what we see with

**[68:38]** offensive and defensive tactics. Uh an

**[68:41]** attacker comes up with a new type of

**[68:43]** scam or attack method and then law

**[68:46]** enforcement or companies try to figure

**[68:48]** out how to get rid of it, right? And

**[68:50]** then they evade that and you have to

**[68:53]** whack that mole down too. And it just

**[68:55]** continually is like whack-a-mole until

**[68:57]** you get to something that's really hard

**[68:59]** to patch or really hard to fix. And then

**[69:01]** it takes a little bit of time. Everybody

**[69:02]** freaks out for about six months and then

**[69:04]** it's patched and then there's another

**[69:05]** one. That's like the way and the cycle

**[69:08]** of security, right?

**[69:09]** >> It's kind of the circle of life

**[69:11]** >> always for everything.

**[69:12]** >> Yeah.

**[69:13]** >> All right. I want to round this whole

**[69:14]** thing out by giving you a quote that I

**[69:18]** heard you say.

**[69:19]** >> Okay.

**[69:21]** >> Cuz I think it's great advice. You use

**[69:24]** the phrase be politely paranoid.

**[69:26]** >> Yeah.

**[69:27]** >> Why should people be politely paranoid?

**[69:30]** Be politely paranoid because the

**[69:33]** information that's out there for almost

**[69:35]** every person on the internet can be used

**[69:37]** to trick you or the people around you.

**[69:41]** Scammers are praying that you don't know

**[69:44]** what scams look like. They are hoping

**[69:46]** that you fall for it, that you fall for

**[69:48]** their urgency, for their pressure

**[69:50]** tactics, for their scheming, their

**[69:52]** spoofing, that you don't know what

**[69:54]** spoofing caller ID is possible. You

**[69:55]** don't know what it looks like. They are

**[69:57]** praying that you don't know what the

**[69:59]** latest scam calls sound like and we just

**[70:01]** demoed them today, right? Yeah.

**[70:02]** >> They hope that you don't know how to

**[70:04]** catch them in the act by using another

**[70:06]** method of communication. And when you

**[70:08]** are politely paranoid, you will catch

**[70:10]** the scammers. So, that's what I

**[70:12]** recommend.

**[70:13]** >> Well, Rachel, I don't think I could have

**[70:14]** said it better myself. Thank you for

**[70:17]** being here. I really appreciate it.

**[70:19]** >> Thanks for having me.

**[70:21]** >> Yeah. Yeah. I knew it. I knew that was

**[70:23]** going to happen. I was like, she's going

**[70:25]** to say it. I knew for sure that that's

**[70:27]** what was going to