2025-03-10 — Agent A.I.

Summary

Podcast Summary: Talkin' About [Infosec] News, Powered by Black Hills Information Security
Episode: 2025-03-10 — Agent A.I.
Release Date: March 12, 2025

---

1. Pre-Show Discussion: Signal’s Presence at South by Southwest

The episode kicks off with Ralph introducing an hour-long presentation from South by Southwest featuring the president of Signal. Although the team hasn't all watched it, Ralph praises the talk, highlighting its comprehensive overview of Signal’s security features and advocacy for privacy-focused communication.

Ralph (00:01): "She has like swagger like she definitely had like I, I would want to go to her one of her talks."

Ralph emphasizes the importance of Signal in promoting freedom of speech and secure communications, recommending listeners watch the presentation for a deeper understanding.

2. The Rise of Agentic AI and Privacy Concerns

Corey shifts the conversation to the burgeoning role of agentic AI—AI agents capable of performing tasks on behalf of users, such as booking reservations and managing communications. The team delves into the privacy implications of granting AI agents extensive access to personal data.

Corey (01:10): "People aren't thinking about AI agents... but at the same time you have to give the agents access to everything."

Jerry adds complexity to the discussion by explaining that agentic AI involves multiple large language models (LLMs) working in tandem to refine outputs, which raises significant privacy concerns as users must share sensitive information.

Jerry (01:40): "Agents are a little bit more complex than just asking the question and being like, oh yeah, look, you did it for me. But from a privacy standpoint, yeah, you gotta give up all that stuff."

The team debates the practicality and current reliability of agentic AI, with skepticism about its ability to handle critical tasks without compromising user privacy.

3. DOJ Charges Against Chinese Hack Groups

The conversation transitions to recent Department of Justice actions, where 12 Chinese nationals affiliated with groups like Isoon and Silk Typhoon have been charged for cyberattacks targeting over 100 U.S. organizations. The team discusses the significance of charging high-level executives alongside technical operatives, highlighting the broader implications for state-sponsored cyber activities.

E (10:16): "It's a good thing. I just thought it was pretty interesting that they charged a bunch of people at once."

Bronwyn notes the DOJ’s incentives for whistleblowing, mentioning substantial rewards for information leading to the identification and arrest of the accused.

Jerry (13:18): "They literally have a reward. We're offering as much as $10 million for information that helps identify any of those accused."

The team views these charges as an encouraging sign of continued governmental efforts to combat cyber threats, despite political challenges.

4. Insider Threat: The Doomsday Device Incident

A significant portion of the episode focuses on an insider threat case where a developer sabotaged his former employer by deploying a malicious Java-based fork bomb. This attack led to resource exhaustion, preventing server operations and deleting user profile files, resulting in severe operational disruptions.

E (30:16): "This was probably relatively easy for them to figure out, but if the user had been more tricky with this, it could get really, really hard."

The team discusses the challenges in detecting such sabotage and the importance of robust operational security (OPSEC) measures to prevent and mitigate insider threats.

5. Vulnerabilities in ESP32 Microcontrollers

Jerry brings up a concerning discovery regarding the ESP32 microcontroller, widely used in IoT devices. Researchers identified undocumented commands that could potentially allow unauthorized firmware modifications, posing a significant security risk if exploited during the supply chain process.

Jerry (34:04): "There is an actual CVE with undocumented codes in this device that supposedly is put in by China to control or modify the firmware at an early stage."

While the vulnerability is currently mitigated by the requirement of physical access to exploit, the team acknowledges the potential for such weaknesses to be leveraged in more sophisticated attacks.

E (35:39): "It could let you create a backdoor... but I don't think it is necessarily a backdoor."

The discussion underscores the broader implications for IoT security, emphasizing the need for vigilance in securing hardware at all stages of deployment.

6. Decline in Cobalt Strike Usage

Corey highlights a notable trend: an 80% decrease in the use of Cobalt Strike for cyberattacks. This decline is attributed to the tool’s detectability and the rise of alternative frameworks like Sliver, which offer more stealthy operations.

Corey (45:35): "Cobalt Strike attacks are down 80% because they're not as good anymore."

The team interprets this trend as attackers evolving to adopt more sophisticated and less detectable tools, reflecting the dynamic nature of cyber threats.

7. Critique of AI-Driven Cyber Attack Surveys

The team critically examines a recent survey claiming that 87% of organizations have faced AI-driven cyber attacks within the past year. They express skepticism about the survey's methodology, sample size, and the vague definition of what constitutes an AI-driven attack.

Corey (53:29): "I think this is poppycock, honestly. I think the data set's wrong..."

Bronwyn and Ralph discuss the potential for such claims to be exaggerated or influenced by marketing tactics, cautioning listeners to interpret such statistics with a healthy dose of skepticism.

Bronwyn (49:07): "It's like, how many devices could you buy that don't have backdoors?"

8. Ransomware Delivered via IP Cameras

A case study is presented where ransomware was deployed through compromised IP cameras. The attackers exploited vulnerabilities in the cameras’ firmware to gain access to internal networks, ultimately deploying ransomware that disrupted operations.

Ralph (55:08): "They got to the webcam and used the webcam to route around EDR."

The team discusses the importance of network segmentation and robust endpoint detection and response (EDR) systems to prevent such lateral movements within networks.

E (57:29): "I mean, 99% of organizations need a pen test. That's a coincidence."

9. Final Remarks and Community Engagement

The episode concludes with the hosts promoting upcoming events and encouraging community participation. They emphasize the importance of continuous learning and collaboration within the infosec community to stay ahead of evolving threats.

---

Notable Quotes:

• Ralph (00:01): ”She has like swagger like she definitely had like I, I would want to go to her one of her talks.”
• Corey (01:10): “People aren't thinking about AI agents... but at the same time you have to give the agents access to everything.”
• Jerry (13:18): “They literally have a reward. We're offering as much as $10 million for information that helps identify any of those accused.”
• E (30:16): “This was probably relatively easy for them to figure out, but if the user had been more tricky with this, it could get really, really hard.”
• Corey (45:35): “Cobalt Strike attacks are down 80% because they're not as good anymore.”
• Corey (53:29): “I think this is poppycock, honestly. I think the data set's wrong...”

---

This episode provides an in-depth exploration of the intersection between AI advancements and cybersecurity, insider threats, vulnerabilities in widely-used hardware, and the evolving landscape of cyber attack methodologies. The team offers critical insights and practical recommendations, making it a valuable listen for professionals seeking to navigate the complex world of information security.

Transcript

[Ralph]

I kind of want to talk about this because it isn't a news source. I, at least it's a news but I don't see it in our news. Do we need to add anybody? Not. No, I don't want to add it because I don't know, it may get interesting. It's more pre show banter talk than it is. It's an hour long video by south by Southwest with the president of Signal. Did anybody else watch that?

[Jerry]

No. Yeah, I don't even have an hour.

[Corey]

To do a lot of things on it though.

[Ralph]

It was, it was very good. She has like swagger like she definitely had like I, I would want to go to her one of her talks. It was highly suggest going to listen to it. It's basically just like the signal 101 and most these people at south by Southwest don't understand security. Right. It's a little bit of a different crowd there. But she's just talking about like why everyone you would should use signal, how they use signal, a little bit of like just examples of people using signal, getting caught doing stuff that like freedom, more freedom of speech stuff rather than like actual illegal things. But if you haven't watched it, I highly suggest it. I'll throw the link, I'll throw the link in.

[Corey]

I don't know if it was the same talk, but she was in the news this morning talking about just basically how people aren't thinking about AI agents. Like when you say like oh hey, like book a, book a reservation at an Italian restaurant that's pet friendly and email everybody on this call line the appointment. Like that's really convenient but at the same time you have to give the agents access to everything.

[Ralph]

And it's like this same, same talk Jerry, about how south by how by an AI agent will have to have complete access to your signal. And then once again someone has your data.

[Jerry]

Another thing too that people don't realize about like how an AI agent works is it's not just a single large language model. Right. You don't just. An AI agent is not just what you see right now with ChatGPT you are, what you're doing is you actually are training the model for your use case. You're giving it a lot of lines. And then what you can do is you have these models feed themselves back. The agent part is you can have the LLMs a bunch a team of them all compare if it fits what you want to do and give ratings and then keep throwing it back over and over again until it's a Good candidate, right. To actually give to you so you can validate stuff. Agents are a little bit more complex than just asking the question and being like, oh yeah, look, you did it for me. But from a privacy standpoint, yeah, you gotta give up all that stuff.

[Ralph]

You give all access. Right. Cause they're talking about how the new AI agent, like Jerry was saying, can pretty much like book you, book you a vacation, then save it to your calendar, save the tickets in your phone, and then message everyone on your signal group that, hey, I booked it. Here's the exact same one you should book too. And she's like, yeah, yeah, right. But it's called the State of Personal Online Security and Confidentiality by south by Southwest Live highly suggest. It was an excellent.

[Jerry]

Interesting.

[G]

That's probably why my talk was so popular because it was all about doing it local. So I had people coming in and they have air gapped win 2k systems that they can't get rid of and they, they want to be able to do stuff. And so local agent.

[E]

Is there a local agentic AI? I don't think there is.

[G]

Agentic.

[Jerry]

Agentic, yeah.

[E]

You mean agentic is the problem. Agentic is the AI that has access to all your stuff.

[Jerry]

Well, so, so there's, there's two things there. Right. So you can still build an agent with an LLM, right? Well you're saying is that this agent already has. So you can still do that, Corey.

[Ralph]

Right.

[Jerry]

You can use like 8NI, which is an automation platform. Excuse me, in 8IN. And what you can do is you can pipe that to do all of that stuff. You can have it. But the thing that ChatGPT has is the ability to move the cursor and mouse around and click on those interactions.

[E]

Yeah, I was gonna say, I mean that's the real concern. It's not about LLMs, it's about agentic AI because it can basically impersonate you, like on all things. That's the whole point. If it can't impersonate you, then it's not useful. Right. Like, which has huge privacy concerns if it can impersonate you. I don't know if there is a locally hosted one or if it has to be cloud based. Honestly don't know.

[Jerry]

Oh, I'm sure there'll be a locally hosted one at some point.

[E]

I mean, I think it's more hypothetical at this point than realistic. Bronwyn's talk was you can do this today and not have the cloud, have all your data.

[Jerry]

I think you could do that for like 95% of the use cases, but the actual one where it moves around and, like, makes it click in your browser. The funny thing is, though, is that web Application all accepts APIs anyway. So that clicking thing, that's for you, the human, not for the computer. Right. So you could do all of those actions without the clicking part if you program that into some other kind of automation.

[G]

So, yeah, Selenium does it all the time.

[Jerry]

Yeah.

[E]

Yeah. So the question is, is there one? I mean, so ChatGPT has the agentic, but it's 200 bucks a month. You have to be on the pro plan. The other funny thing, the reviews I've seen of it have been like, this is hot garbage.

[Jerry]

I try it. It's. It's at least right now silly. It's kind of silly like.

[E]

Like, I haven't seen an agent I would trust to book my travel who would do that.

[Corey]

Right?

[Ralph]

Like, the rear, middle seat in front of the bathroom is where Corey's sitting now, thanks to this new agent.

[E]

AI or just like, I found a more ideal route. It routes you through Dubai. Because you said two years ago, you said you wanted to go to Dubai. I just read this signal message from two years ago when you were like, I'd love to go there. So I just went ahead and added that to your itinerary.

[Bronwyn]

Is that any different than using the airline program straight up? Like, I just, I downloaded the. The app from the airline and it sent me through Dubai too. So.

[Jerry]

But the thing about these, it must.

[E]

Hate you because I've never had that happen to me. Oh.

[Bronwyn]

I've been routed weird places just to get from, like, Minneapolis to Milwaukee. It's sort of like Minneapolis to Milwaukee by like, the most convoluted method possible.

[G]

That's an airline issue.

[Bronwyn]

That's an airline issue. Yeah.

[G]

That's not an AI issue.

[Ralph]

That's a U issue. Why did you click and accept that fight?

[E]

I will say, though, are we being honest? AI isn't the ones making the decision about where to route people we don't know. I will say I do think I can barely figure out how to do it. And not to, like, toot my own horn, but I think I'm a little bit better at like, using a web apps than AI is. So I'm like, I don't know if I would trust AI to, like, do critical functions of my life, like booking travel or doctor's appointments or things personally, but that's just me.

[Ralph]

Jerry, how do you survive on that fast of Internet?

[G]

Yeah, seriously.

[Ralph]

I literally, I literally bought the Most expensive Internet I could in my area because of my upload was so slow when I was doing simply Defensive that.

[E]

Like, most people don't have an option.

[Ralph]

Yeah, well, it's a gig down and 250 up.

[Corey]

I think the problem is my children are home from school, and, like, my youngest is an iPad kid. He's got, like, nine devices going simultaneously. He looks like the keyboard is from sticks.

[Jerry]

He's like, one of those, like, factories in China where they have, like, a million phones.

[E]

He's got, like, the big stack of, like, 10 phones.

[Corey]

He's like, yeah, he's running Tiger Pot Farm.

[E]

Hello, and welcome to Black Hills Information securities. Talking about news. It's March 10th. What's up, everyone? I wasn't here last week, so I'm assuming this podcast has now been converted into an AI only podcast, Is that right?

[Jerry]

It's actually the number one for the AI we can only talk about AI is.

[E]

Are all of you replaced with AI copies of yourself? And I'm the one being, like, the schizophrenic one?

[Jerry]

Yeah.

[Bronwyn]

My instructions don't allow me to answer that.

[E]

Should that be a reality TV show where, like, that, you know, the person is, like, talking, you know, with a bunch of people and they're all just AI, don't they?

[Ralph]

They. They pretty much have that with the Catfish show, right? Isn't there one where, like, people are trying to cat, like, you just throw an AI in there and no one will notice? Probably.

[E]

Goodness.

[Jerry]

Did you see the video where two so. So someone called, and it was like, an AI assistant, and they said they were an AI assistant. And then they had the other phone that was like, I'm an AI assistant, too. They're like, all right, let's change our mode of talk to, like, a beeping because.

[Ralph]

Gibberish. Gibberlink language.

[Jerry]

And then they start beeping at each other.

[E]

Wait, is that real? Is that real?

[Ralph]

It's real. It's a real form of speech. Yeah.

[Bronwyn]

And they even did that in Terminator 0 to where it's sort of like, oh, you're a fellow infiltrator. Let's just talk.

[Jerry]

Where do you get the idea from? It's always.

[Corey]

Yeah. The cool thing is, too, the first thing it did was transmit a public key.

[E]

Wait, a public key or a private key?

[Corey]

Well, it did a public key to the one unit, and the other unit replied back with its private key encrypted with the public key so they could exchange.

[Jerry]

Now they have a.

[E]

So it's more secure than I. Anything that humans can do.

[Corey]

Yeah.

[Jerry]

Yeah.

[E]

That's even more terrifying. Although I guess you could argue that biometrically identifying a human is like, we're evolutionarily pretty good at that. Like, face recognition, voice recognition, gate recognition. Like, it. It's actually crazy. I mean, some people have face blindness, but, like, I'll notice people at. Especially, like, when I'm looking. Like conferences or airports or whatever, but, like, just by how they walk or things. I'm like, wow, my brain's pretty good at this. Good thing. I don't know who this person is at all, but I know that I know them, so it's pretty cool.

[Jerry]

Wait, you know I'm doing that at online?

[E]

Yes. Yeah, that's fair. It's harder. It's harder online. But I think I'd know if someone replaced you with a fake person, Ralph. I think I would know.

[Jerry]

Oh, yeah. Well, this is AI, so you.

[E]

You're not.

[G]

I have a hard time believing that any AI could capture his Persona, personality. Most of the AIs are still pretty flat when it comes to personality.

[E]

I understand.

[Ralph]

I'm screwed.

[E]

I agree. I think. I think AI would more work in a situation where people don't know each other. It's kind of like you were talking about catfishing. Yeah, like, I think it would. It works better when it's people, you know that don't. Aren't familiar with each other.

[Jerry]

We got some spice. So we want to talk about.

[E]

We got. Yeah, I mean, there's some spice in here. So, I mean, I think that you guys want to start with the US treasury stuff. Chinese hackers charged that. You want to start with that one? Start with some laying down the law.

[Jerry]

I thought you were going to go.

[E]

Right to the AI story.

[Jerry]

We already died for a minute.

[E]

Yeah.

[Corey]

Okay.

[Jerry]

We have to get out of it.

[E]

All right. I think this is an uplifting news article, in my opinion. And it's depressing and uplifting at the same time. But so basically, the Department of Justice charged on. When was this article posted? March 5th. So five days ago. They had charged 12 Chinese nationals with attacks targeting more than 100 U.S. organizations. There's like, a few different groups of attackers that they charged. They charged some employees of isoon, which we've talked about isoon on the show before. They're basically a pen testing firm, but with most of their contracts paid for by the Chinese government. And they do all kinds of sketchy stuff. They. The reason we talked about them before is because someone leaked a bunch of their internal stuff on GitHub. So we know A lot about how they operate, what they sell, all that good stuff. But there's also Silk Typhoon, which is an apt. A couple people were charged. Interesting thing about the people that were charged from Isoon, which is a company, is it includes like the executives, right? So it's like the CEO, the CFO or whatever, and some like, technical people. So it's kind of interesting.

[Ralph]

That is a bit of a different whammy. Usually it's just all the techie nerds, right? And they just paste their pictures up to actual do. A CEO who may actually travel somewhere where he could be extradited is definitely a lot cooler.

[E]

And I mean, it's basically saying this company's bad, right? It's like they also charge the sales director. Like even, I guess, selling their services, they're calling that as bad. I don't know. I just thought it was pretty interesting that they charged a bunch of people at once. I also see as uplifting that the doj with the current political thing is still charging people. I call that a good thing. So, you know, we're still doing. Still doing stuff. So that's good.

[G]

I think that's more a matter. They just had momentum.

[E]

I'm still counting it. There's a lot of things.

[Jerry]

We need some good.

[E]

I'm still counting it as a win. That's what I'm saying. But yeah. So, I mean, obviously I will say they charge people and then they're like, we don't know anything about these people.

[Jerry]

If you have anything, you give you $10 million.

[E]

Yeah, we'll give you like, they literally have a reward. We're offering as much as $10 million for information that helps identify any of those accused. So I guess they're kind of encouraging people to flip. I mean, that's a big reward, right? $10 million?

[Jerry]

Yeah, $10 million. You literally could probably bring them in a black bag and they'd be like 50 bucks.

[Ralph]

2 million for any information leading. The 2 million for information leading to the arrest is probably the bigger one.

[E]

Well, that. No. Well, so it says information that helps it identify any of those accused of carrying, directing, or carrying out isun's malicious cyber activity. Basically saying, who actually did this work? I think they just want more information. But then the 2 million is for the apt. For the Salt Typhoon or whatever they are, or Silk Typhoon. Sorry. Yeah, So I. I don't know. That's a lot of money. I guess they're trying to encourage people's loyalty to maybe shift a little bit. Have you ever thought about traveling outside of China.

[Corey]

I feel like 10 million is like the most DOJ can ever offer. I've seen it a couple times. I've never seen it north of $10 million. And it's kind of funny that it's like a cap. I mean, I guess it is a federal business or a federal agency, so they can't just go Hamilton. But not to fully change stories, but that tech company that got its crypto wallet hacked the other day and they're offering $140 million.

[E]

Yeah, but it was like a billion.

[Jerry]

Dollars that got taken, though, so. Perspectives?

[Corey]

Yeah, it's 10% of the amount taken. But it's just insane how very much, like, hold my beer. But I mean, yeah, $10 million is no joke. I do think it's like one of those ones where, like, if they get convicted and they do, like, it's like the $10 million is like a series of milestones that have to occur that.

[Jerry]

Will take years to get to.

[Corey]

But yeah, getting them out of the country, that's number one, step one.

[E]

Good luck taking that money or using it. If you are based in mainland China.

[Jerry]

Yeah.

[E]

You're going to have to travel to take advantage of that money. I mean, I guess we can do the AI one. What was the art. It was basically. I mean, I guess for those that skipped the pre show, which is totally fair, it kind of made the news. There was a talk by the CEO of Signal. Right. Is that. Is that correct? Which I was like, okay, first of all, why does Signal have a CEO?

[Bronwyn]

How.

[Corey]

Or the president? I think her official.

[Ralph]

She's the president. She's the president.

[E]

Yeah, dude, she's president.

[Ralph]

She's pretty awesome. You need to listen to that whole talk.

[E]

Like, but she's like, who's paying her salary? I'm not.

[Jerry]

It's a foundation. There's a Signal foundation. You could donate. Yeah, okay. You could donate. And it's a yes. And then people do donations to. Not. Not to small ones, big ones.

[Ralph]

The person. The person that's interviewing her too, is like, why should I donate? And she's just like going through it. And he's like, well, how much can I donate? Can I subscribe? And she's like, no, you go in settings and there's a donate button. Just push it and put a bunch of zeros and we'll be good.

[Jerry]

It's.

[Ralph]

It's a really good time.

[E]

Yeah. So it's a super awesome talk. It's AI. So, you know, you've probably gotten enough of that lately. But I mean, I think the biggest thing that we were discussing in the pre show is just the privacy implications. So they're talking about agentic AI specifically, which, for those that don't know, agentic AI is essentially the agent part is the AI can function on your behalf, it can impersonate you on various places. And the use case we were talking about in the pre show was you want someone to book your flights for you, but you're, you know, you can't afford an assistant or, you know, a person who would actually do that. So you have chat GPT. Hey, chat GPT. I need a flight to Austin, Texas on February, you know, whatever. And it would just book it and it would also, you know, put it on your calendar and send it to your friends that you booked it and all that good stuff. Obviously there's huge convenience benefits in theory. Although I will say, like, we were joking about the reality of that, but the privacy concerns are huge. Right? Like that. That was basically the majority of the talk, right? Or am I miscarric?

[Ralph]

No, yeah, that is a majority of the talk. Just privacy. And how signal doesn't. Doesn't take or store any type of data and how it's completely open source so you can tell and verify that it doesn't take any data. Right. I also didn't know a lot of encryption messaging software uses some of like a lot of signals.

[E]

Protocols.

[Ralph]

Yeah, signals protocols, which I thought was pretty cool, right?

[E]

I mean, signal is really hot right now. I feel like just in the world, like, so didn't the government say, like, use signal?

[Ralph]

Like, Jerry's finger over there is itching.

[Corey]

I know I won't play sound effects on someone else's show, but.

[E]

Yeah, no, feel free.

[Ralph]

1. One of the interesting facts she actually said is like, I think it was all of the, like the Netherlands, I believe it was, has the highest level of signal saturation. And it. I want to say it was like 90%. It was something that was amazing that like half the, like most of the country uses signal, which was pretty cool. Someone can fact check me real quick and go look at that video. But how do you get your mom on signal? Right?

[Jerry]

Go AI.

[E]

I know I've been trying to convince my friends to do it, but like, for now we're still. There's a couple of like Luddite holdouts in my friend group that like keep everyone on rcs.

[Jerry]

It's actually really simple to set up signal too. Like it takes like a minute as long as you have an iPhone to takes nothing.

[E]

So, yeah, the only real. The only Real argument against it is it's another chat app.

[Jerry]

Yeah, another chat app. Yeah, another app I need installed.

[E]

That's the only real argument against it. But I mean, I don't know. People, people are always like, isn't that what criminals use? I'm like, I don't know, you know what criminals use? Freaking texting.

[Jerry]

Like, I don't know what.

[G]

Criminals use. Pagers.

[Jerry]

Yeah, no, according to all the books I've read, they definitely are into these encrypted phones and other things like that. But in general, they obviously they, they want to not have their text read in the, in the courtroom after, you know, all of this evidence comes out.

[E]

So, yeah, basically, if you are, and I will say, like we're also during the pre show talking about, is agentic AI actually good? And in my opinion, I don't think it is yet. Maybe it will be someday, but right now I would not trust. ChatGPT has a $200 a month subscription where you can access Agentic AI. And I mean, Ralph, you said you tried it. Was it useful? Did you, did you have it broken plates for you?

[G]

No, no.

[Jerry]

So I wouldn't use it in that way. But another way to do agentic AI would be another service that you sign up for, which, by the way, is going to have the exact same problems. But the way it would work is you would say, hey, I want to book a flight, right, And I want to get the best price. And you like, kind of give it the details and they'll be like, all right, well, give me your credit card, give me these things. And it won't. Yeah, no, I hear you. Give me, you know, give me access to your calendar on Gmail or whatever or. And what it would do most likely is it would look at all those flights all the time, trying to make sure it was getting the best deal. And when it made a match, right, Using AI agents to, to do this, you know, in a schedule, when it made a match of what you were looking for, it would send you a notification being like a yes or no question. Do you want to book this or not?

[E]

Right? Here's the deal.

[Jerry]

This is, this is what we got. If you say yes, we'll automatically book it. And that's kind of the other idea of an agent. What the ChatGPT's version of it is. You literally kind of say the same thing. You're like, hey, I want to book a flight for this day and blah, blah, blah. And what it does is it opens up a browser and a virtual machine and then has the AI server Selenium.

[E]

It's freaking Selenium all the way down. It's popsicle sticks and glue. But is it everything? I mean, I don't know. It'd be one thing if there was APIs and things, but like, do we really think, like, random major airlines have functional APIs?

[Jerry]

And that's the point. That's the point of it is to be like, hey, we could do anything in the browser because we're not reliant on learning the APIs. Right?

[E]

It's like, have you heard of a captcha?

[Jerry]

Oh, oh, gotcha.

[Ralph]

I'm not a human.

[Jerry]

I am a human. I think captchas are kind of silly.

[Bronwyn]

Yeah, I've seen people build these Both using the ChatGPT one and the offline one. And in both instances, like, it would have a problem with looping through the commands being like, okay, find a server and buy this server for us. And I'll be like, oh, okay, well, I ran into an error, so I'm just going to loop back. And then like, before you know it, you have like 408 servers like that.

[Jerry]

Added to your card. It's like.

[Bronwyn]

Book me a trip to, you know, Texas. That is like, okay, cool. I bought the. I bought the entire plane just to be safe.

[E]

We didn't know preferences. Like, you were unclear any of the.

[Bronwyn]

Parameters about you have the entire plane. Thanks for giving me a number, by the way.

[E]

If there was like a mode that was like planes, trains and automobiles mode, where it would just book you the most like, ridiculous combination of different things. It's like, all right, step one, you take this U Haul, you drive it to Milwaukee. Then you get on this train. You have to illegally get on a boxcar that's headed for Portland. Then halfway through, you get off in Des Moines, Iowa.

[Jerry]

Then you're have to treat agent AIs just like you would treat a actual assistant. You wouldn't just say, here's my credit card. Do whatever you want. Whatever flight, don't ask me, I don't care. Obviously you do care. Obviously you should get asked. And so, you know, you should treat them the same way.

[E]

Yeah, it is funny. Basically, we're not there yet. We still can't spell strawberry correctly on most LLMs. So I don't know if we'll be having things booking our flights anytime soon, but either way, it is a privacy concern. I think most people would be a little bit. When you get to the part where it's like, okay, to assist with your request, I will need. And then the list of things is like 80 things long. There's going to be an AI that just gives you all, that holds on to all your private data and gives it to other AIs. Like, I mean it needs your credit card number, your address, your name. Like I think people would be a.

[G]

Little I to do that. I don't need.

[E]

Yeah, but every time, well, every time you ask the agent to get to do anything, you have to give it all that information. So you better have another AI that talks to the.

[Jerry]

Yeah, it's a whole thing.

[G]

An AI gapped AI.

[E]

Yeah.

[Corey]

I don't know how far you want to go down this. I've seen some interesting structures where you have an oversight AI agenic community that has Socrates and some of these thinkers. And then beneath that is your interface AI, like your actual assistant. And then under that are purpose built AIs and you dispatch to the, your assistant and the assistant dispatches to the individual functional AI. People like this one just sends email invites, this one just books flights or whatever, you know what I mean? Or this one only does restaurant reservations. And then they start tasking around with each other. And it's actually kind of cool if you see it Complicated. I agree. There's a lot of, it's fraught with problems. And if there is a problem where you know, it falls down at one point, I don't know how it gracefully exits. But that's kind of the, that's kind of the, the strategy I've seen with some of these agents. It's like a bank of agents where you have purpose built functional utility. It almost seems overkill. But I mean, who cares? You spend up more agents.

[E]

So it's like subscription TV where the, the goal is to save money on TV and then end up spending more money long term because you have like 80 different subscriptions. It's like, well, yes, you already have airline AI, but do you have hotel AI? And you also do car.

[Jerry]

And the only reason you got it was to save a little bit of money booking a airfare. But it costs you more for the AI than you actually save in the airfare.

[E]

Yeah. Also everyone's doing it. Well, there's that. And also then airlines are going to start being like, listen, our websites are going down constantly. If you just book with the app, we will give you 50% off.

[Jerry]

Exactly.

[E]

Like, for God's sake, just freaking book. Yeah. No, I mean, I don't, I don't.

[G]

Know that I want my AIs to go down the same road as phone apps. I mean, I Can't even keep track of how many dozens of phone apps I have.

[E]

You only have.

[Jerry]

If you want to talk about something going down today. Do you guys see that X got DDOSed into orbit?

[Ralph]

I did see that. I was told it was, and then it was never down for me, though.

[Jerry]

Oh, it was down for me for a while.

[G]

It was down for me for a while.

[Jerry]

Yeah.

[Ralph]

Maybe it was just an east coast thing.

[Jerry]

Yeah, I got the message. I was like, is it down? Is it really down?

[E]

It's really down. Is there a confirmed article of them getting ddos or did they just screw up some code?

[Jerry]

So, yeah, they literally just screwed up. No, there are multiple articles and Elon claims a massive cyber attack and.

[E]

Okay, but that's Elon. What actually happened?

[Jerry]

Well, we don't know what actually happened.

[G]

Less than massive cyber attack is not sufficient to fluff his ego either.

[E]

Wants to be a.

[Jerry]

The website went down now, why? We don't know, but it sounds like.

[E]

It was a DDoS apparently dark storm claim. Does anyone know Dark Storm There was.

[Jerry]

Claiming that they had done this, so. Yeah, So I couldn't access croc.

[E]

I don't know how I'm going to.

[Jerry]

Live my life because there's not 15 other eyes, but whatever.

[E]

Oh, you just got to use another AI to access it.

[Jerry]

I know.

[E]

Yeah. I mean, DDoS is. Don't generally make. You know. I mean, it is what it is. Like, it's. Everyone's susceptible to DDoS.

[Jerry]

Yeah. All banks are susceptible to bank runs. All websites are susceptible to ddos.

[Bronwyn]

Yeah. Well, and then as for the attribution, like, I've seen actually kind of a couple different places that have been giving the attribution. So. Yeah, you have Newsweek saying that it's Dark Storm. You have Anonymous saying that it was down.

[Jerry]

Yeah.

[E]

So, yeah, but they just claim everything, right?

[Bronwyn]

Yeah. It's going to be the same messy attribution when you're still in the middle of everything being under siege.

[Ralph]

So they're a pro Palestinian cyber group.

[Bronwyn]

Yeah.

[E]

Maybe all those servers they shut off and got rid of had some purpose.

[Jerry]

Right. I don't know. They saved money.

[E]

So there's a couple articles in here about like, a doomsday switch that an IT employee or someone put a developer put into their software does any little. Kill switch.

[G]

Kill switch.

[Corey]

The.

[E]

The summary of this is pretty hilarious. Does anyone want to go through this?

[Jerry]

Sure, sure.

[Corey]

Let it ride.

[Bronwyn]

Yeah.

[E]

All right. So, I mean, I don't know if this is, like, based on the actual data, but, like, I don't know what, what to call it? What I call it the flavor text. Like the, the summary that gets pulled in by notion is. Yeah, it's pretty funny. I don't know if you guys can see it or Ryan, if you can share it. Look at the summary there where it says, is David Liu enabled an active directory? Not anymore. Is David Liu guilty? Yes. Is David Lou facing jail also? Yes. It's so funny.

[Bronwyn]

So that was a function he set up as part of the code for. It's like, you know, is. Is DL. I mean this is DL enabled in Active Directory.

[E]

I mean it's so transparently hilarious. Like he literally made the function. Is David Liu enabled in Active Directory? Like was it AI? Like, I don't know. Like that seems so.

[Corey]

They'll never figure it out, I guess.

[E]

Hubris. I don't even know how. Like you gotta have a better system than this. I mean, I guess the.

[Bronwyn]

And I was gonna, I was gonna do a similar thing being like, come on, you gotta have better like defensive techniques than this. But this happened about six years ago. Five, six years ago. 2019. That's sort of like 2019.

[E]

What? OPSEC didn't exist in 2019.

[Bronwyn]

It did exist. It just didn't have as many great tools foreseeing somebody doing this level of access and turning off other people's access in Active Directory or within a bunch of systems ago, something would have mapped through all of that.

[Ralph]

But depending where he saved this, this would be really hard to find. Right. Like he's a developer, he has access, he's going to be able to go anywhere.

[Bronwyn]

Yeah.

[Ralph]

Usually like when he gives his two weeks, that'll get put on. Hopefully he gets thrown on some type of list. Right. But it would take a while in order to find something like this, at least for a defensive person. I'm not going to be looking around for kill switches or like.

[E]

But can we talk about like, what if he just forgot his password? Does he like, if he just forgets his password?

[Jerry]

Well, so that's a different one though. So that would be locked, not disabled.

[Bronwyn]

Disabled. Yeah.

[E]

Okay, I guess. What if he goes on extended leave? I don't know. Anyway, it's just funny, you know, if.

[Bronwyn]

They turn him off an active directory and this might have been one of the things where it just, you know, permissions mapping and going. It's like this, you know, this dev has the permissions to do what now? And you know, you terminate him and double check that, you know, those permissions. That sounds like something that is a bit more possible today. But Five years ago might not have been something that a lot of organizations wrap their heads around.

[E]

Yeah. And well, I mean I will say you're right that it came before and we've talked about several articles on the show of like employee sabotage of ex employers. There have been a ton of convictions and crime and sentences handed out for this. Probably most of them after 2019. I feel like now the writing is pretty clearly on the wall that if you do this you will be get in trouble and you will like there's been multiple articles but maybe in 2019 people were less aware of that. Another fun tidbit is like it was all done in Java and it was basically like a, from what I can tell, kind of like a Java fork bomb which basically says creating more and more non terminating threads. It would consume resources until the code crashed, preventing people from using the machine and then also possibly deleted some files or throw away other people's files.

[Ralph]

If you're gonna do it, go hard. Not just like a little thing. Right. Like he's going to jail. He might should have like, like come on. If you're gonna make RM-RF every domain.

[Jerry]

There'S ransomware with no ransom credit encrypting everything with that I will say though.

[E]

I think you'd get. Would you get more than 10 years for that? I don't know. Well, at least if he, if he.

[Ralph]

Ransomed it, they would probably lose logs. They would lose logs at least. And they may not be able to track it back to him, but with this they're still able to track it back to him.

[Jerry]

Like yeah, they could have thought they got hacked just coincidentally when he got terminated.

[E]

I mean, I guess if you don't want to go to jail, don't build a doomsday device against your company. But if you are going to build a doomsday device, make it a doomsday device, not just a what's going on with these three servers device.

[Jerry]

Yeah, it's got to cost them hundreds of thousands of dollars, but whatever, I'd.

[E]

Be surprised if they actually get 10 years. That seems that was.

[Jerry]

That seems like for a first offense. And it's mostly property damage, right? Yeah, yeah.

[E]

And like I don't know, I guess it'll probably depend on the actual impact. Of course the company will argue that it was the biggest impact that's ever happened. But I mean if you're relying on java not to DDoS you in the first place, I mean, come on.

[Bronwyn]

So let's see looking through a little more bit more of the details on this one, it seems like it might not have been a permissions space, but it deleted files that were associated with user profiles, so then those user profiles just couldn't load. And then the servers got caught into, like, a boot loop and just kind of kept crashing. He had been, like an indirect lockout. Yeah. He understood the system.

[Jerry]

If he gets 10 years for this, then CrowdStrike could get like, a million.

[Bronwyn]

Oh, yeah.

[E]

Well, okay.

[Bronwyn]

Just causing an infinite boot loop. Yep.

[E]

Fair. But I mean, like, basically, I guess the moral of the story is people who have intimate knowledge of how your business works.

[Jerry]

Yeah. Insider threat's real.

[E]

I mean, insider threat. Yes. People who have intimate knowledge of how your business and code works can do a ton of damage to you in a way that can be really hard to figure out. This was probably relatively easy for them to figure out, but if the user had been more tricky with this, it could get really, really hard.

[Jerry]

You guys want to talk about this? The ESP32s and.

[E]

Yes, please.

[Jerry]

Yes. So.

[Corey]

So hot.

[Ralph]

All right, let Jerry talk about it. Jerry, go for it.

[Jerry]

I was going to minimize it. I'll let. I'll let you say it, and I'll say.

[Corey]

No, no, no. I want to talk about it because I do. I really do want other perspectives on.

[E]

So, I mean.

[Corey]

Yeah, go ahead.

[Jerry]

Yeah, I was just going to say. All right, so here's the. Here's the big story, and then we can talk more about, like, the. The what and the. How it really affects. So the ESP32, its microcontroller on all kinds of other devices all over the world. There's thousands or millions of them.

[E]

1 billion billion units with a B.

[Jerry]

They're very, very popular. They're very popular for a couple different reasons. One of them is they're extremely low on power. They're extremely low on cost, and they're very versatile. Right. But supposably. And I said, supposedly there is an actual CVE with undocumented codes in this device that supposedly is put in by China to control or modify the firmware at an early stage. Okay, so this is undocumented. And what the attack would be is that in either a supply chain, other things like that, you could compromise this device early and be able to then compromise the device that it's used in later on in the supply. Right.

[E]

It's Bluetooth attacks. There's. There are.

[Jerry]

There is some, but most of it is a physical attack on the device.

[Bronwyn]

Yeah, yeah.

[Jerry]

So the idea was, is that if you compromised it in process, you could do whatever you Want later on you could have any kind of message sent to the device to take a further compromise, but that you would need physical access to the device to initiate these, you know, low level commands that were undocumented to essentially root the device.

[E]

So the concern is that these functions allow pretty easily a backdoor to be created on these devices.

[Bronwyn]

Is it a backdoor though? Is it a backdoor? Is it just, you know, you can control from code that's running on the ESP32?

[E]

Well, it could let you create a backdoor. I don't think it is necessarily a backdoor.

[Jerry]

I think the, no, I think, I think the argument was, is that it was undocumented and they didn't acknowledge that you could do this. So they were saying that you shouldn't be able to do this. So that's kind of like the backdoor idea where this code or these, these commands were still in the device, they were in the chip, but no one was talking about them. So once they found them, now it's like, oh, yeah, well, of course you could do that and totally take over the device, but you're also bypassing what should be some kind of trusted security controls.

[G]

Well, they're also claiming that those commands are debug commands. And yeah, it's an update. If you scroll down and you see an update for today, the company is claiming that their debug commands used for testing purposes. And of course, nothing bad has ever happened from a developer leaving debug code or, or debugging turned on.

[E]

Right? Yeah, I mean, it could go either way. Like, it's hard to say whether it's actual, like genuinely mistaken debug commands or if it's intentional. Right. Like, who knows? How cynical do you want to be? Your call, at the end of the.

[G]

Day, whether it's deliberate or not, it's still a vulnerability and somebody malicious is going to take advantage of it.

[Jerry]

Is it, is. It's a, it's an undocumented way to root the device. But, but, but anytime that you have physical access to a device. Game joke, right? Like, that's, that's like kind of the, that's like the thing, right? I mean, every defcon, they're like, oh, well, you got physical access to it, so now you own it.

[E]

Right.

[Jerry]

Like it's like, it's, that's kind of the concept. Now. There are things that, you know, are supposed to be secure enclaves or whatever that is supposed to protect from physical access. I'm just saying, usually once you have physical access, it's kind of a, you know, I mean, I'll just take the damn ROM right off the board now.

[E]

Right, okay, that is true. But I think the doomsday scenario here, you know, to go back to the last article would be. Let's say I'm. Let's, let's make it personal. Let's make this personal. Let's say I'm publishing a bunch of badges for a conference. Maybe a wild west hack infest conference. Maybe those badges, Maybe those use ESP32s, which by the way they do. Let's say hypothetically I get all these devices, then I put some kind of weird backdoor into that.

[Jerry]

You've ruined every CTF now for my.

[E]

Own purchase, for my own, like I do this for myself, then once I have my backdoor I can then exploit it remotely, right? So like it essentially gives me like persistent long term access to my own things that I have the control over, right?

[Jerry]

Yes.

[E]

So I feel like the dooms. The real threat, the real threat here isn't I bought this device and now someone can exploit it like to make a botnet or whatever. It's more like a very specific scenario like we're getting into like Mr. Robot type stuff here of like, okay, I bought a sketchy vacuum that uses an ESP32 or whatever and then my other sketchy vacuum that uses an Android based system is going to exploit my system that uses ESP32. You know what I mean? It's like these weird like knock on effects of, you know, potential vulnerabilities. I will say though, kudos to the researchers. I mean this is such a cool vulnerability to find and undocumented commands are not that uncommon. But this is still really awesome research.

[Bronwyn]

But it doesn't seem like they open up these devices to third parties for building onto them. So there's also that there's the comparison to like the intel hardware as well. Like intel has tons of hardware that has undocumented features on their processors.

[E]

Why?

[Bronwyn]

Because you're supposed to use intel drivers on the intel processors. It's not opened up to third party and it's like this express if like it's not opened up to the third party. So even in like the scenarios where you go, okay, if I take this thing and I load third party coding onto the radio, then I can do bad things, which is a couple of hops for ifs that you still see that in the other things to where it's like if I manage to load unsupported drivers, third party non intel drivers onto intel hardware I can leverage undocumented features of that intel hardware.

[E]

I mean, maybe I'm confused, but I'm pretty sure if you buy an ESP32, you can write your own firmware to it, right?

[Jerry]

That's how it works.

[E]

I mean, that's the whole point.

[Jerry]

Yeah, that's the whole point. Yeah, you're writing your own firmware.

[E]

So like, I'm not sure if that necessarily is super true, but I don't know.

[Jerry]

The vulnerability too is found in the Bluetooth side of the chip since there's a bunch of. It's like a little socket. So it was found in the Bluetooth, but it, from what I read, it doesn't appear that you can remotely attack this just via Bluetooth.

[Ralph]

Yeah, this is lame.

[Jerry]

Then you can compromise it in advance and then do that.

[Ralph]

Get out of here. Get this. Let me hear what Jerry has to say about this and then get this out of here. If you have to have physical access to it, I'm like, if it was remote.

[Corey]

No, I agree. I just wanted to talk because, listen, I think using security researchers on staff as a marketing tool is excellent because it benefits the entire community. And we see this all the time. Like you see Microsoft researchers and, you know, crowdstrike researchers. And it's really, really cool because we as practitioners can appreciate it. And it's better than like a frigging, you know, funded webinar about whatever, identity governance or something. But. But when the security researchers, the tech people who are doing the cool research and find this and then they're going to go to Rooted Con and the marketing team gets a hold of it and literally publicly comes out and says tar Logic Security has detected a backdoor in the ESP32. And then bleeping computer is a news outlet picks up that verbiage and throws it in the headline to say a backdoor, which they've since walked back and changed it to remove the keyword backdoor in the title of the storyline. It really does caused this since like this almost like uptick sensationalism. And again, I don't blame the security researchers. I think it's cool what they do, but because it's been like not weaponized but like utilizes this marketing arrangement. You have these people who are not understanding what it is that they're actually reporting on and it just gets whipped up. And now like we're having this conversation about literally a nothing burger type story that is, you know, like probably going to run on the news because it's everywhere. You know, it's China, you Know what I mean?

[Jerry]

So you said the logic word bingo.

[Bronwyn]

Yeah.

[G]

Are we sure that an AI didn't write the article?

[E]

No, we're not sure. I mean, I guess what I would say is like, the funny thing is like, it's more just burying the lead about Iot security. Right? Because if we're being honest, like, how many devices could you buy that don't have backdoors? Maybe, I mean, we don't know. But personally, in my life, I just assume every IOT device I buy has a backdoor.

[Ralph]

My Bluetooth shovel is completely locked down.

[E]

Whether that. Now, whether that backdoor is a feature to me that, you know, maybe changes.

[Jerry]

But I think low level back doors and mainly just total lack of security on the IoT device is probably way more common.

[E]

Right, like default creds like that.

[Jerry]

Exactly, yeah.

[E]

You know, like, of all the devices out there, the ESP32 is probably actually pretty high up on security. You know, like, it's probably more secure than your average IoT device. Or at least, you know, devices that use it are probably more secure. But, you know, and it gets into like the. Oh, well, this is why we rolled our own. Well, did you, you know, do you think it's as tested as the ESP32 at this point? Probably not. It's a whole thing. But I mean, is there a greater, like, should, you know, is this a concern? Like, probably not.

[Jerry]

It would be a big concern if you could remotely attack devices without having to do anything. I'd be like, holy crap, this is wild. Right? But if you have to, you know, get on the thing to compromise it, then I'm like, yeah, it's like, it.

[G]

Makes it a lot less of an issue.

I mean, it's one of those things that we see not just with physical, but with, oh my God, this is a CVSS 9.9 vulnerability. But to get it to there, you need to go through 10 different steps and two of them require this special access and one requires this special access and this one requires you to be elevated.

[E]

And.

[G]

And so much in our industry gets sensationalized that this is just another cog in the wheel as far as I'm concerned.

[E]

I will say for the record, it's only a 6.8 CV.

[Bronwyn]

Yeah. And just like for a lot of vulnerabilities you say, well, is there a metasploit module for it? Like for this Bluetooth thing? I'm wondering, like, is there a flipper module for it?

[E]

Nope.

[Bronwyn]

And no, there isn't. So if it was something to where you Go, oh, you can break Bluetooth and do all sorts of horrible things with it. Then it's like, okay, well then, you know, where, where's the, where's the flipper module for it? Which that might come, but it's not there yet. So I think again, kind of a.

[E]

The flipper module is just a little thing that says step one, obtain physical access to the device and upload a backdoored firmware. Yeah, it's like, yeah, that's basically it as you go.

[Bronwyn]

It's like, okay, you, you upload some compromised firmware and then you can do weird things with your Bluetooth module on your flipper. Okay.

[E]

Step one, commit a crime by stealing someone else's personal property. Step two, exploits available, like. Yeah, I mean, anyway, what else has people got? Jerry, you got anything spicy? What else is hot right now?

[Corey]

No, I mean that, that Bluetooth ESP1 was the one I wanted. I can certainly, you know, dig something up in a hot second.

[E]

Cobalt Strike up some hot second, dig up some hot articles.

[Jerry]

What happened with Cobalt Strike?

[Corey]

So Cobalt Strike attacks are down 80%. So.

[E]

Wow.

[Jerry]

It's. Because it's not as good anymore.

[Corey]

Well, I mean, that's the story. I mean, Timmy, I said this, this morning. I said it doesn't mean there's a. Any degradation in the amount of threat actors using post exploitation frameworks. It just means they're, they've moved on to sliver or something else. Because Cobalt Strike is easy to detect if you don't like tune it up.

[E]

It's also kind of hasn't really been updated very much since it was acquired. So.

[Jerry]

No, it's kind of. It's a dying C2. I mean, not to say like all the C2s are very tough to keep up, but you know.

[E]

Yeah, I mean, I also feel like at this point attackers are like C2. But I have quick assist, why would I even bother? Yeah, well, I know what article we could talk about. So the. Did you guys see the one that was like the majority of orgs have been hit by AI cyber attacks.

[G]

No sensationalism there.

[Jerry]

Yes. And I honestly that this article, because I just looked at this for you, was written by AI to talk about AI. It's like a self controlled.

[E]

So I will say I don't, I don't want to disparage this company too much. What are they called? Safe. Safeco or so Safe. So Safe Safe. What was the company Safe. Hold on, what was the company from. From Mr. Robot? All safe or something?

[Jerry]

All Safe.

[Ralph]

All Safe was.

[E]

Yeah. So all safe. Definitely not. Evil Corp has. I. I went to their website and tried to download the actual report. So. And they wouldn't let me without a business email. And at that point I was like, nah, I'm good. And it made it to you, Corey. So basically they good luck getting your hand on this. Hands on this report without giving up your work email, which I'm not going to do. But anyway, this is survey based reporting. Okay, it's survey based. So I just want to outline that first. But essentially 87% of security professionals have reported that their organization has encountered an AI driven cyber attack in the last year. How do they know that's the headlining? What does that even mean? Does anyone want to take that off?

[Jerry]

Does that mean because it's really bad and it had like mistakes or does it mean because it was really good and it got access.

[Ralph]

Grammar and phishing emails has gone up. That's why everyone's using Grammarly now.

[E]

I will, I will survey. Wade. Wade, has your organization encountered an AI driven cyber attack in the last year?

[Ralph]

Without a doubt, yes, if you count phishing. Right. But you think I'm, you think I'm comparing all the phishing emails. You're crazy.

[E]

How could you.

[Jerry]

You can have AI do that.

[E]

So that's.

[Jerry]

Did I write this? Tell me.

[E]

That's exactly the point I wanted to make is how would you know who is.

[Ralph]

Maybe they're getting, maybe they're getting the fancy Alex, like voiceover, like things where people are taking voices and then being like, hey, this is your manager.

[E]

87% of companies have been targeted with that.

[Bronwyn]

I don't know about the 87%, but yes, it's like I've, you know, I've dealt with these for both the voice and video items that now you currently are deep fakes. Yeah, yeah. The deep fakes that you're receiving and from this article, because it's like I've even, It's like I did a small talk on my AI evolving the threat intel. Like this article, like how I can tell it was written by AI is that it does not give any sort of suggestions for countermeasures. Like, well, what do you do about this then?

[E]

Why are so safe?

[Bronwyn]

But a lot of it comes down to like your user awareness training now because you're still like in my organization. Even like still right now. People go, how do I know it's a phishing email? Because it has like spelling and grammar errors that, you know, your security awareness team put in. It's like, stop training your users to look for that. Or AI is just going to like stomp all over because AI is just going to write things that don't have spelling and grammatical errors. So if that's the only way your end users know to identify phishing emails, you're going to need to update that. And you know, what's the whole. It's like, you know, and then the, you know, you know, the dismissive, like, oh, you know, people aren't faking voices, faking the CEO's voices and sending phone calls. Yeah, they are like, you need to, you need to make your, your staff, anybody that, that receives phone calls and might have something that they're in a position to initiate wire transfers, stuff like that, they need to be aware of this saying, like, hey, there is the possibility that you in the wire transfer room receive a phone call from a fake, the fake voice of somebody with the authorization telling you like, hey, yeah, we need to pay this invoice. Like send this, you know, just over. It's like it. Well, it sounds like Bill and this sounds like something Bill would want us to do. So yeah, sure, let's send the money. So, okay, your end users need to know these are attacks that are possible and not just edge cases and something that's being, you know, over dramatized from Mr. Robot. Like these are things that are actually happening out there.

[E]

I mean, multiple hilarious jokes in the discord. One most orgs hit with AI attacks. Yeah, adsense. Also another hilarious joke is like, does it count if you use Grammarly to write your phishing email? Does that count as an AI?

[Corey]

It does.

[Ralph]

It does, right? I will tell you this right now.

[Jerry]

They're hip hop compliant though.

[Ralph]

My HOA has received AI submitted forms because any form I submit to HR to hoa, I put through AI. So it's, it knows their HOA rules. And I say, hey, write me up, write me up a submission for new slider glass doors that meet the HOA requirements and boom, there you go.

[E]

I'm going to do the same thing for having an hoa. But I will say, like, I think that the thing that really bugs me about this article is they say AI driven. AI driven. I think that's where I'm caught, like catching an issue. It's not like there's this master robot AI that's like hacking a bunch of organizations. Someone is hitting enter on a keyboard. This is not like they're using everywhere.

[Jerry]

It's everywhere now. It's going to be in everything.

[G]

Are they confusing a botnet with AI?

[E]

No, no. I think they're just. It should, it shouldn't be AI driven. It should be like AI influenced or AI assisted because AI is not the one executing the attack. They're just using AI in their attacks. I don't know.

[G]

I think what bugs me the most about this article is that it's only 500 security professionals and 100 customers. So what, 600 people is the survey that's extrapolated out. I don't think that's a big enough sample size for them to be able to even say something like this.

[Jerry]

That's a perfect size.

[Bronwyn]

Yeah.

[E]

And I, I mean there's no, it's not like every organization has a procedure for identifying if the attack is driven by AI too good.

[Jerry]

It must be AI. That's the only way we could get attacked now.

[E]

It's like saying it's a massive cyber attack when your website goes down.

[Jerry]

Yeah. And so that we can say, and the insurance is now going to have like a new rule that's like, you know, like we don't pay if it was AI.

[E]

Let's hope not.

[Jerry]

Anyways.

[E]

I mean, I definitely hope that. I mean, I don't know. Jerry, what is your take? Give us your take on this one. It better be.

[Corey]

Yeah, I mean, I think it's, I think it's poppycock, honestly. I think, I think the data set's wrong, like to say they, you interviewed 150 people, 500 people, and 130 of them said they have high confidence that they can detect. It's, it's later on in the article, but 26% said that they have high confidence that they can tell when it's an AI attack or not. Like, things like that are like, was this survey done at a, like a corporate happy hour or something? You know what I mean? Like, like, I love, I love reports, I love metrics, I love, you know, objective, you know, evidence to support arguments. But this just is salacious. It's like, like this is a, this is a classic example of like, I could tell you without question I know what so Safe sells just based on this report that they found. You know what I mean?

[Jerry]

Like anytime you see a company, what do they sell?

[Corey]

Or like a company releasing a, like a report that like phishing emails are up 7,000% and this report has been funded by insert email security gateway company. You know, it's, it's classic. So like this is, this is, you know, to me, this is like something that you read in the shopping grocery line. Like, right.

[E]

Like natural sun and it's like Inquirer of Cybersecurity.

[Ralph]

Sun is bright.

[Corey]

I think about it.

[E]

I think you nailed it. I mean, I will say we are about to publish a survey that says 99% of organizations need a pen test. That's a coincidence.

[Jerry]

The data supports it. We asked everybody that called us, we.

[E]

Asked all of our clients if they need a pen test and they said yes. No further questions.

[Jerry]

Got our numbers.

[E]

It makes sense.

[Jerry]

Come on.

[E]

No further questions. 100% of companies need it. We asked everyone, they said yes. And by the way, yes, we got everyone drunk first.

[Jerry]

Oh, God. Well, that's kind of the marketing. So I think that's the, the idea, right? That.

[E]

So, you know, the other. I think. I know it's, you know, we have three minutes, so this is going to be a. I guess we have technically seven minutes, but.

[Jerry]

Ooh.

[E]

So, okay, did everyone see the ransomware? Who deployed their ransomware from a webcam or from an IP camera?

[Jerry]

Yeah.

[Ralph]

Yeah. Did you read the full report?

[Jerry]

Compromised an IP camera and then used.

[E]

I didn't read the full report.

[Ralph]

I. I was a little confused by it at first because I'm like, well, how is this working? Like, why does the webcam matter at all with this ransomware? So they originally got it, just let.

[E]

Some work around Ed.

[Ralph]

They, they pretty much weren't able to get the ransomware into the network because the EDR was blocking them. So they got to the webcam and used the webcam to route around edr. But shouldn't the EDR detect the ransomware when it's moving from webcam to computer?

[Corey]

So my understanding, because I covered this article, my understanding was that the, the webcam was running obviously a flavor of Linux and it had. The file server had SMB enabled. So it used SMB to get to the files on the router and then deploy the ransomware, I guess, from the webcam, doing like a Linux variant of it and then wrote it to the file server with the Linux.

[Jerry]

How did they execute it then? Because wouldn't.

[E]

They executed it from the webcam.

[Ralph]

They executed it on the webcam. The webcam was hitting stuff and then.

[Jerry]

Rewriting, Reading the data, reading the share.

[Ralph]

And rewriting the file ransomware. That's exactly what I was thinking.

[E]

Well, I will say, Ralph, you know, some of these, I mean, I don't know if it was an NVR or actually one of the cameras itself. I don't.

[Jerry]

I feel like it had to be came in there. Is that, Is that.

[E]

Am I. No, but I feel like it had to be an NVR Yeah, I think.

[Jerry]

It'S probably an IP camera because those do definitely have, you know, a full operating system and they have a lot of different protocols and they would absolutely support SMB natively in that firmware. So you wouldn't even need to bring it in Linux.

[E]

It wasn't firmware. I mean, I guess, but that's what I'm saying.

[Jerry]

I'm like in, in that little Linux distribution, right, that's running on this camera. It would already support SMB because that'd be a common protocol in the. They call them firmwares, but I mean, essentially in the operating way, it would be easy to go from there to the next. Right?

[E]

So, yeah, basically. Long story short, it's a cool, it's real world demonstration of something we've warned companies about for years, which is, I mean, two things. Number one, make sure you have EDR and everything. Now putting EDR in your IP cameras. If you do that, you're awesome and I love you. Wow.

[Ralph]

Wow.

[E]

Second of all, you know, I have.

[Jerry]

A. I have a recommendation. Maybe you should isolate all your cameras onto a totally separate network that doesn't touch, because I don't know why anybody needs to be.

[Ralph]

Stop making smart choices. All right? You guys sell IR service services too. You need everyone to have those.

[Jerry]

I'm sorry, I'm sorry. I'm making a recommendation around the things.

[E]

Yeah, no, I mean exactly, like segment your IP cameras. It's cool, though. I mean, it's scary.

[Jerry]

It is definitely a way of like actual hacking where you're, you're given a limitation. You're like, all right, let's figure out a way around it. Because I can't make this payload any better because I bought it from, you know, ransomware as a service. So anyways.

[E]

Well, let's also talk about how they would have detected it if they just had looked at the EDR alerts.

[Ralph]

Yeah, exactly. That's what I was gonna say.

[E]

The initial access came, but I mean, basically it tells you you shouldn't. You should be scared because your network's flat, but you shouldn't be that scared because they would have detected this.

[Ralph]

I really, I really hope someone just didn't get the alert and was like, oh, it deleted it. Like, we're good.

[Jerry]

Oh, yeah.

[Ralph]

Where did it come from?

[Jerry]

That, that's, that's tomorrow problem. And this is.

[E]

Is this Friday night?

[Ralph]

Is this the same one? The same breach? Because I know recently, I can't remember where I heard. I want to say I heard it from Jerry that a threat actor ordered Like Uber eats or something to the actual workplace of their target to see that if they were there or not. Is this a different one or is that the same one? Anybody else hear about that, or am I just. Did I dream it?

[Corey]

I don't think that's the same attack.

[Ralph]

It's not. Okay. Yeah.

[Corey]

And I don't know if I told you about that, but I vaguely remember this. I think it was DoorDash. I mean, it was Doordash matters, but.

[Jerry]

Yeah, they totally have worse kids.

[E]

So what? Like, I don't. Is it like a target, as in a company? Like, they're just like, is anyone there?

[Ralph]

Yeah, they're like, hey, go to the. Like, order this food. And then. And then, like, in the instructions, because you can actually communicate with the doordash people. They're like, tell me if anyone's there. And they do. And they're like, hey, nobody's here. Like, you're like, we just left the food at the door. Like, okay, cool. Now we deploy the ransomware because we know nobody. No one's there. That was my thought because they're deploying it via this IP camera. It's going to take so long.

[Jerry]

They're using DoorDash as a scout, as a surveillance. Yeah, yeah, because they're in a different country, so they obviously don't have.

[E]

Like, can you just call the company's main phone line?

[Ralph]

Why? This is way cooler. All right? This is.

[E]

Gets us to talk about a lot of stuff. What do you think they ordered? Like, one Twix bar or something? Like, is it the lamest doordash order? What are they sending them?

[Jerry]

Ordering food to go find out information. Like, hey, was there anyone wearing this outfit over there?

[E]

Yes. Okay, so I'm just gonna. Let's go around the room. You're ordering doordash to a target company you're about to ransomware. What do you send them food wise, Wade?

[G]

Pizza.

[Ralph]

Why are you do me last? I was trying to think of something good.

[Corey]

Wade's sending in and out all day long.

[Ralph]

No, in and out. In and out. Doesn't Doordash. How dare you? I would know better. I was trying to think of something like the local. The local, like, Russian restaurant, like some goulash or something like.

[E]

Yeah, yeah. Oh, pierogies or something.

[G]

The scary thing about the recon by DoorDash is that could be used for a whole lot of nasty physical types of things for stalking, whatnot.

Yeah.

[Jerry]

There's stories of people, like, take 10 photos and send them to me.

[E]

Here's the doordash instructions. Okay, step one, go to Best Buy, purchase a laptop, install this thing called Kali Linux. Then install this thing called Eve. It's all this thing called Evil Twin. Put the laptop in your backpack while it's still.

[Jerry]

I'm good at doing really good.

[Bronwyn]

Yeah, I mean, you just have it delivered in like a pizza box there. You just have it be like the seventh pizza box down. Is your, your thing with Kali Linux or you just order. You just order and deliver them like some Chinese food and you just get. You get like a little waterproof like box, put it in the bottom of like the grossest Chinese food you can get in a container. You get that delivered. They're going to be chowing down on like all the yummy Chinese food. And then you just have like this one that like, I don't even know what it is that people don't eat of Chinese food.

[Jerry]

Sure.

[E]

I mean of the company. And be like, will you compromise your account if I send you free food? And they'll probably.

[Bronwyn]

It's probably just easier.

[Corey]

What do you want?

[Bronwyn]

You know, here's a 50 gift card. Tell me your password.

[Jerry]

How about licensed human iTunes gift card?

[E]

Oh, nice. Any last thoughts, articles? Concerns chicken wing related follow up article?

[Ralph]

Next time I'll bring a chicken article. I promise I'll find something I will say if you are attending. Besides San Diego, we sold out of tickets by the way, so very nice. We. We should. I'm trying. We're trying to find another way to get some more tickets in, but Bronwyn's class still has a couple seats available for it, so it's like a secret.

[G]

Way learn how to install local LLMs. Yep, yep.

[Ralph]

She had a standing room only in in at Scale, which is another Linux here in Los Angeles.

[G]

It was amazing. Well, scale's always a fun time. And it was. It was really, really nice to get to hang out with people that I don't get in real life very often. And I was floored by how many people showed up and how much interest there was in doing local LLMs. And that was very cool.

[E]

All right.

[G]

I also want to add in if you're Interested in B sides312 the day after ThoughtCon in Chicago. Our CFP closes on Friday and our tickets are available now.

[E]

Cool.

[Corey]

Awesome. Wild West Hack infest CFP is open.

[E]

That's true.

[Jerry]

Yep.

[Ralph]

Don't tell people that. I have to then compete with more people. All right.

[E]

He's not gonna get the speaker again.

[Jerry]

All right.

[E]

All righty. Thanks.

[Jerry]

Appreciate it. Yeah, later. Thanks for showing up.

[Corey]

Midnight Jump. The shark. Buddy, we'll have to talk. That new song.

[Bronwyn]

Okay. No, I. I know. I get.