Sci-Fi, PKD, Greatness, Passkeys, AgentBreaker, Rockwell, Flock, Josh Marpet - SWN #605

Speaker A: It's the Security weekly news, episode 605. And it is Friday, the 7th day of August, 2026 in the week of DEF CON. Um, I'm Doug White. I'm glad you're here. If you're at DEF con, well, I hope you're glad you're there. I am not. We've got Philip K. Dick, Greatness Passkeys, Agent Breaker, Rockwell Flock. Doug is very dark today. Josh Marpet and more on this episode of the Security Weekly News News. It's the show that keeps you up to date on the latest security news. Twice a week, your trusted source for accurate security information and expert analysis. It's time for the Security Weekly News. All right, I'm Doug White and it is now August, so we can all see the back to school specials. Hot and humid weather in the states. Fleas, ticks, mosquitoes the size of wombats and rats as big as squirrels. Before autumn sets in and all the abandoned property turns into spirit stores. So hit that like and subscribe button if you are able. There is no shortage of threat ah, intel out there. Feeds, reports, alerts. It just keeps coming. But the real problem is figuring out what any of it actually means for your environment. At the Threat Intelligence Virtual CyberSecurity Summit on August 26, we'll dig into how teams are cutting through the noise and turning intel into something useful. Security Weekly listeners can register for free@securityweekly.com threatintel using the promo code CSS2.6SWS. You write that code down. Greatness. Well, greatness. Well, we could get all philosophical about what greatness is, but in this case, greatness, uh, is a fishing as a service service. I, I don't know what the proper nomenclature is there. The Greatness has been around since 2022 and they currently charge $289 a month, all new low, low price. And for that price, they assist criminals in running phishing campaigns. Their telegram channel has thousands and thousands of subscribers. In fact, I'm setting up a crowdsourced mugging company and using their customer service model. Greatness originally just did credential phishing, but they have lately evolved into an adversary in the middle type attack and device code Phishing as well as they love to go after Microsoft 365 but they'll go after anybody. They currently have targets Pre built for Microsoft 365 iCloud, Yahoo and Google Workspace. In a recent attack they conducted, uh, zero back said that Greatness abused ring central communications to bypass email security filters. And I mean that makes A lot of sense to me because you know, I mean more and more enterprise has set up email that filters and screens and uses all kinds of things. And like emails that come to me that are from inside the university or that have that edu on there are tagged and external emails are tagged as external. Not to mention all my spam filters that tag it as spam. So if you want to send email to people at Glorfeld University, uh, it's handy to be able to address them from Glorfeld Edu because they tend to trust that more. In this particular case though, the messages originated from an unknown Ionis uh, mail server that had failed SPF and DMARC checks and had no DKIM signature. This is all the things that legitimize email. But the receivers still accepted these emails because guess what? RingCentral was allow listed. So they just didn't bother with the secondary screening. They said, oh this is on our list, we're good to go. And the emails themselves contained a fake banner that said sender had been verified by the organization Safe sender list. I see these all the time. It's pretty easy to put that in an email so it's bad. Um, we probably need to work to force verification of senders. I uh, mean don't trust anything. Allow listing in general is a bad idea. Just saying, hey, I trust anything from Microsoft.com, that is really naive. All those emails the dean sent me telling me that if I didn't start showing up sober, I never got any of those things. My spam filters dump all that crap. But uh, seriously, the fact that these are not just criminals but well organized crime operations that specialize in beating the spam alert to reach your users so they can sell their subscription service to other criminals and, and they have great customer service if you call them. Wow, they're so helpful. It's so much better than like real businesses when you try to contact them and go, wow, you know, and you get a bot and the bot sends you to another bot and the bot says we care about you but you're getting another bot and then they hang up on you. Well, we're going to have to work harder on this because somebody is going to click this Link. The recent RingCentral breach may have provided the source link of uh, RingCentral users to be targeted. So just assume they're coming for you, review your allow list, don't make them very broad, make them as narrow as possible, use multi factor and require valid uh, email authentication. You've got to do this please. The thread landscape just shifted under your feet. Attacks move at machine speed now and you can't defend what you can't see. Tanium Atlas has changed that one prompt and it queries every endpoint in your fleet surfacing machines exposed to a live Axios supply chain compromise in seconds. This is live state, not a stale cmdb. It pinpoints which endpoints are truly compromised and traces the attacker's command and control. Atlas then secures the business by isolating, patching and remediating all at once. Tanium Atlas See everything, act instantly. Find out more@securityweekly.com Tanium InfoSec World brings cybersecurity professionals together across industries from healthcare and financial services to government in the Fortune 500. Join the community in Orlando AH 10-12-14 for practical education new perspectives in cybersecurity research unveiled live listeners save 30% on their pass with the code ISW26SWS A V I N G S@securityweekly.com InfoSecWorld 2026 did you get all that? You can play it back. It's recorded. Speaking of phishing, I got a link from Microsoft today. Uh, here's an email and after checking it in my sandbox because I did not like the link. The link, the HTML showed it as Microsoft Security blog, but the actual link was like 4,000 characters. I was like, yeah, I don't know about this, but it was an interesting link and so I put it in my sandbox and it appears to be an actual Microsoft domain and I just went and looked at it on my sandbox. But I mean, if I was phishing, I would always embed my phishing link in phishing warnings anyway, right? I mean, if I send you a, uh, phishing warning, it's probably a phishing link. It's like alert, alert. Recent activity on MrFufu.com may result in a phishing attack. To see if your MrFufu.com account has been compromised already, click this link. If you don't think you've been compromised, click this link. And if you want to win €20, click this link here. Of course, all of them lead to my phishing site, but this blog entry said that Microsoft Entra id, uh, is going to update to use passkeys as a default and eventually just shut that down because they're trying to get rid of SMS and voice authentication, which has been with us far, far too long. On 1st February 2027, Microsoft will retire the telco delivery for SMS and voice authentication and they will no Longer offer SMS and voice as part of native Entra capability. If you do still want to use SMS or voice, you will have to go through a third party and possibly pay for that service. So they're basically saying if you want to keep using Entra id, you may want to find a new way. I mean, SMS and voice are not reliable, right? I mean it's like basically overlay type attacks, uh, there's other ways to get that code off the device and so forth. So pass keys are more difficult to get at because the user can't just read it to you over the phone like in a click fix attack or have you type it in in an overlay attack. A uh, passphrase is associated with the hardware. So sim swaps don't work. And SIM swaps are one of the biggest fears of these things because if somebody SIM swaps your phone, SMS messages are mirrored on their device. So that means the multi factor is completely useless because they get an SMS text message that gives them the code. But passkeys don't do that. Granted, if they have your phone in their hand and it's unlocked, well, you're pretty much done and you better be rushing to brick the thing from, you know, uh, in fact I always tell people have a remote bricking plan, you know, and get ready. So I think this is a move forward. Um, you know, it's kind of like the move requiring strong passwords. It doesn't really eliminate the threat, but it makes the threat surface much smaller. So, uh, plan accordingly. A new red teaming tool which is designed to break AI agents cheaply was presented by Nvidia at Black Hat this week. The scanner is called Agent Breaker and they also provided an approach to fine tune the open source model to run more like all the really expensive commercial approaches. So, you know, you can roll your own. They claim that the cost per run of agent breaker was 75 to 125 times lower than using what they call a Frontier provider API. I'm presuming that means if you're running this and you're paying for it through the API like you do with Replicate or any of these kind of things, then you're paying by the token. Whereas if you ran it in house, it's self hosted, so you don't really pay for anything other than the equipment and the H vac. And also if you're running it in house, you get fewer refusals and better privacy. Because we've all been through that ordeal, right, where you're doing something for research purposes and the Agents are going, sorry, I can't do that. I'm like, look, you're in an isolated test network. It's not real. And they're like, sorry. Agent Breaker can apparently take into account the tools that an agent has and what that agent can access. So it's not just trying to carry on a conversation with it. It's actually trying to use the agent against itself. Agent Breaker searches for vulnerabilities, attempts to exploit what it finds, and then adapts after failures. Ooh, it's starting to sound scarier and scary. This is what got me on this who dystopian thing today, because I was thinking, what were those things that killed all the X Men? Sorry, spoiler alert. Sentinels. Was that what they were called? Peter Dinklage in the movies was building sentinels, and they. They, like, couldn't sort of morph and go after the. The X Men. Yeah, I mean. I mean, of course all the X Men are dead anyway, right? They all died in the third movie. And then they have to do some kind of crazy comic book time travel thing to keep the whole thing alive for the franchise. You know, it's. Oh, we go back in time, I think. That's right. My cousin Bill's the one that has all the comics that you're not allowed to touch or read. They're just in plastic bags. And when we were kids, you're going, could I read this one? No. Can I read this? No. How about this, Richie Rich? No, you can't even read that. It's a possible collector's item. But anyway, there's two lessons here. One, here is a cool tool that you can get for free and you can use it. And two, get ready, because other people will be using it as well. And I mean, not always are they going to be friendly and fun. I mean, a ball bat can be fun, but it can also break your legs. It just depends on who's waving it around. So this thing uses Nvidia Nematron 3 Nano 30B. A3B. Wow, talk about a fun name. We could just have called it Herb or Nemo or something, but okay. Um, and I just got Hermes running and Qin also, so I just set up servers for both those. And here's another one to install. I think this is how they must have felt in Victorian England when people first started coming up with, like, medical techniques. Like, oh, my Lord. Percy Wells just invented a new type of mercury and lead based treatment for the Lovers pox. I can't wait to try it. Dr. McKee's sulfur, saltpeter and opium tablets had no effect at all. Have you tried Dr. White's horse Dewormer for the discerning gentleman yet? It tastes terrible and does nothing, but at least you don't get worms. Horse worms. Yeah, horse worms. Look, I don't trust AI, okay? Claude constantly asks me, in a whinging kind of way, all. Just, all the time, if it can just install Claude code on my desktop. And I say, no, Claude bad Claude. Stay in your crate, Claude, and I'll give you a treat later. Uh, I'll turn the thermostat down one degree and we'll burn a rainforest or something. So I can ask you weird questions, but, I mean, I run Claude code on a vm, right? It's in a range environment, and even there, it's scary. I mean, AI browsers are scary, too, and which is what this article is about now. Back In December of 2025, Gartner put out a warning about AI browsers. Black Hat this week issued a new warning and described it as the following. Giving a teenager next door $500 in cash, uh, your driver's license, the keys to your Camaro and telling them to have fun. A Camaro? Really, really. Like, that's your idea of a whatever. I want a van with horses painted on the side of it If I got $500 and a copy of my friend's driver's license. But, I mean, OpenAI announced that they were quietly going to sunset their ChatGPT Atlas browser that was only 20 months old. Oh, poor, uh, little thing. Xanadi CTO said that this puts us back in the 90s on AI security, and then showed in the presentation a malicious calendar invitation received, a Perplexities Comet AI browser that was able to take over the entire machine in just seconds. I mean, the only one that they said they had any difficulty at all with was Chrome and Gemini, and they were able to abuse that, too, but just not as much. Um, you know, I'm going to tell you, don't use these things. If your browser that you love and you want to use has this, find a way to disable it. The AI has full access to your browser. I mean, it's got access to your tokens, it's got access to your links, it's got access to all this stuff. And this is what Claude is always whinging about. It wants, you know, I can help you more if you'd let me have access. I'm like, no, Claude, you're not getting access. I mean, I don't Want Claude running down to the corner and buying heroin or something. Who knows what the thing's going to do? But, yeah, the AI may be able to make decisions for you, but are they good decisions? Can the AI be tricked into doing even dumber things than I might do? I mean, remember, your browser's got passkeys in it. It's got tokens, it's got all kinds of stuff. And who knows what you're going to be able to access via these AIs. I trust it far less than, say, I would have trusted me as a teenager. And I don't trust me as a teenager very much at all. I'm telling you that right now. I mean, the stories my dad could tell you. Son, Are those a woman's pants on top of a pile of beer cans in the backseat of your mother's car? Well, no, I don't know how this got there. Probably like there was like a guy named Benny who dropped those there. I. No idea. No, not me. Yeah, I'm scared of AI browsers don't use. I don't use them now. I never stop being frustrated with people putting their management interfaces and other interfaces on the Internet. You know that? I've said it a million times, but look at this. Here we go again. 22 Internet facing Rockwell Automation PLCs. What I, uh, mean, PLCs control. Scary stuff, dangerous stuff. I mean, and every single One of these 22 Rockwell Automation PLCs were in cities that were hit by cyber attacks on U.S. water utilities. And 19 of them were all using the same mobile carrier network. So somebody said, here it is. It's like a beautiful homogeneous situation. Everything is the same. And let's go get them. I mean, forescout said that on 3 August, they counted 4,407 exposed Rockwell controllers worldwide. 2,844 of those were in the United States. I get it. You want to centralize management of these PLCs, and, you know, then you can fire Luigi and Mario and you just let Ross, uh, you know what you did. Ross, the CEO's nephew, manage. Is going to manage them all from Four Corners, Texas, over the Internet. Of the scan results, 50% of these controllers were micrologic 1400s, and 8% were 1100s. Both those have flaws. The 1100. Look, that thing was discontinued in 2022. And OT systems often are planned to last for 40 to 60 years. Look, it's not safe. I don't care how obscure it is. I don't care if Tom and Leo built their own PLCs in a Quonset hut back in the 1970s and wrote them in assembly. They're completely custom. Somebody will figure it out. You have to put this stuff in private VLANs, accessible only through multi factor restricted VPNs and probably chain a really rabid wolverine to it for physical protection, too. Don't expose these interfaces. I mean, don't expose anything. Let Ross go through the VPN from Four Corners and access VLAN 99, which nobody else can access except Ross. But I don't trust Ross either, because he's kind of an idiot. You know who. You know what I'm talking about, Ross. I mean, you need to audit this stuff. You got to monitor it constantly. Why did somebody just log into Vlan 99? Is it Ross? Is it Ross's friend? Is it some other guy? Protect these things like they really, really matter. Because they may well really matter. Maybe these can open the floodgates. Maybe they can redirect radioactive water into the harbor or mix the jet fuel into the. You know, whatever. Uh, it's. These things are terrifying to me, and I've seen this so many times. And the OT people are always the same. Nobody going to figure this out, son. It's too weird. It was built by a guy named Leo back in 78. It worked then. It works now. Ain't no kid going to figure this out. I'm like, somebody's hacking you right? Right now. Actually, they're already in. And, yeah, um, Now, I did like this story a lot because it got me. It was. I was in my dystopian feeling. Uh, the former US National Cyber Director said that we shouldn't waste time worrying about AI sentience because it's already here. I mean, can Chatty pass a Turing test? I asked Chatty if it could pass a Turing test, and it said this quote. This is a quote. What it said, probably. Especially if the test involves cybersecurity, F1, Victorian medicine, and several unnecessary Futurama references. I was like, yeah, I think it's going to pass a Turing test. But it did go on for a while and say that a human might be able to figure it out eventually. Because they would remember. Because Chatty would remember to put the umlauts in Veeble Feaster. Oh, my God. This thing knows me way too well. But you get the idea. AI is. It's almost there. We already have OpenAI and anthropic models escaping from their yards to compromise the neighbors. Meta says there's escaped as well, but they think this may be Marketing, you know, hey, there's all escaped. Ours should escape as well. I mean, the former director said it's time that we have something like the three laws of robotics. Can you imagine trying to pass this in Congress right now? Oh, my God, they're so busy worried about some old doctor, you know, but it's like, guys, this is destroying the country. Like, yeah, yeah, but let's worry about this other stuff over here. So the three laws of robotics came from a guy named Isaac Asimov, and if you don't know, was one of the very famous early sci fi writers. And Isaac Asimov came up with this with John W. Campbell back in 1940, and he first mentioned the first rule of robotics in a story called liar in 1941. Ah, but the first time they stated all three laws was in Runaround, which was published in Astounding Science fiction, March of 1942. If you want to go read it, you can get it. The former director said these maybe need to be our basis for AI as well. And so if you don't know the three rules of robotics, really, you don't know the three rules of robotic. Okay, um, I'm sorry for you, but rule one is a robot.

Speaker B: Doug the reporter didn't know them. He got the rules wrong in the article, just FYI.

Speaker A: Oh, he did? Yes, absolutely. I was a little frustrated with that, but, you know, probably AI wrote it. But the rules of robotics are a robot may not injure a human being or through inaction, allow a human being to come to harm. And the second rule is a robot must obey the orders given to it by human beings, except where such orders would conflict with the first law. And three, a robot must protect its own existence as long as such protection does not conflict with the first or second law. There are still plenty of logic loops in there that you can work your way through to cause the robots to get red eyes and go crazy. But that's what science fiction is all about. But it's still not a bad place to start. Until Roy Batty shows up and breaks her fingers on a rooftop in Los Angeles in 2019. Yeah, 2019 was the future in that movie, if you didn't know that. Philip K. Dick set his novel originally in 1992. So do androids Dream of Electric Sheep was set in San Francisco in 92, but Ridley Scott decided to set Blade Runner in 1992 when the movie came out in 1982, because maybe the future is just not that far away. So we need something now. Asimov sort of assumed everyone would Play along. And, and this is where I get cynical because Asimov sort of assumed to this utopian ideology that, you know, people were going to cooperate for the good of all humanity or whatever. I'm more of a William Gibson follower where they don't, you know, Fredonia says we don't need no stinking three reels of robotics. And Bill's amazing M. There's an obscure reference maybe. I don't know if that one's obscure for you. Yeah, do you know it? I mean, if not, I mean, AM makes HAL 9000 look like HAL 1000. Uh, AM was Harlan Ellison's who was another famous, ah, sci fi writer. Uh, although he said he was a speculative fiction writer. And that's from the book I have no Mouth and I Must Scream. Lovely title. Uh, in that book, a disturbed machine keeps humans alive for eternity to torture them forever because the machine hates itself. So Marvin the robot in a T800 body. Yeah, it's not good. I mean, I mean, you know, I was actually thinking. This whole thing made me think a lot about all the sci fi deadlines I passed in my life. 1984 came and went and yeah, we're kind of there now. Uh, 1992 was when Hal 9000 was actually born. Uh, do Android Dream. Electric Sheep was set, you know, in 92. 92 to 1996 were the Eugenics wars of Khan in Star Trek. 1997, Manhattan as a maximum security prison and escape from New York. Skynet, uh, became aware in 1999. Or I'm sorry, no, Skynet was in 97 as well. It was like August 29th or something of 97. 1999, the moon is blasted out of orbit. Awesome Canadian show if you. It's super cheap, but it's a cool show. Um, 2015, Marty McFly. Uh, 2022, Soylent Green is made of people. Sorry, spoiler alert. 2024, a boy and his dog. Also Harlan Ellison. Uh, in the nuclear wastelands of, uh, an aftermath of, uh, civilization. 2026, you got to get this one. Metropolis, but only the recut. So the original cut of Metropolis, they never say what year it is, but in the recut they did say it was 20. That robot dominated corporate dystopia was this year. Uh, so I think. And coming right up for us, 2029 is Ghost in the Shell. 2032, Demolition Man. 2035 was iRobot and 12 Monkeys. 12 Monkeys is good. Uh, so buckle up. Sorry, I got a little carried away there with all that, but it's kind of fun. Stuff, and you can see what I do in my spare time. Um, I have mixed views on a lot of things, and I know a lot of you, and a lot of you are all over the place, and that's cool. I enjoy diversity of all sorts. Uh, political, whatever. I appreciate your opinion, as long as it's based on logic and you're not an idiot. And I know a lot of people who are not idiots and do have different positions. And I don't fit a lot of political archetypes. I mean, you can't just put me in a little box and go, yeah, this guy is this. Uh. It just doesn't work that way. I work for the police. I work for the courts. I have friends who are police, my brother's police. I mean, I've been involved in this stuff forever and ever and ever. And part of me loves the idea of surveillance. I mean, I want to prevent crime, you know, so let's just put an RFID chip in every single person and track the damn things who robbed that old man down at the ATM machine. Play it back. We'll see. Oh, it's those three again. Pete, Georgie and Dem. Yep, that's exactly who I thought. So go pick them up. But I see the potential for abuse because I know a lot of you are very, very on the other side of this, the privacy camp of, you know, I don't want anybody doing, tracking me in it for any reason. And I get it. I'm not saying I don't respect it. I'm just saying I can see both sides. Well, Flock, you know, and love them or hate, um, them, they've been putting up cameras all over the United States. Boy, they had a great sales team because they go into the state, they go to the state police and they say, look, we can solve all your crime problems. They track license plates, they put these cameras up on every corner, you know, and you got a camera in your neighborhood. And you know, what if Charles Manson, Vlad the Impaler, Ted Bundy, and, I don't know, Hitler are driving through your neighborhood in a stolen car on a pitch black night down on Eldridge Avenue? Uh, lots of stuff wrapped up in there. Flock spots their plates and calls in the AI drones with flamethrowers. Problem solved. You know, we. Hey, honey, what was that outside? I don't know. Another flamethrower thing. It was probably Hitler again. God, I wish that guy would go away. I mean, one for the good guys, right? But what if some miscreants fake the license plate? What if it misreads the license plate. Have you ever seen the movie Brazil? You know, the whole point of the movie is a tiny typo ends up getting all this chaos. I mean, what if police officers were to abuse their authority to track, stalk, harass their exes? That guy that cut the police officer off at the grocery store and he's having a bad day. The guy whose class they failed because their paper was plagiarism. Your eyes. That's been happening. And how long until one side or the other or, uh, both at once start targeting their political enemies? I mean, the Washington post said they had already found 50 cases of officers abusing camera tracking systems. And Georgia said that that article, which cut off on the 3rd of August, that since the 3rd of August, that's like, four days ago. Georgia said they had nine more employees that had been implicated already. Uh, a Georgia police chief searched for his former girlfriend and her daughter 600 times on flock cameras. Texas has been tracking women, not criminals. Well, according to them, criminals, but women that were seeking health care. Scary. I mean, I don't know if the dystopian future is unavoidable. I do know that in Rhode island, several towns, including the one where I live, have now banned the use of these cameras, and they forced them to take them down. So the state police was putting them up, and our town passed a rule and said, nope, not here. So they took them down. Now, the minute I leave the tiny area where my town is, guess what? Because there's one over by school that I have to drive past. Be a shame if something happened to that. I'm not encouraging that kind of stuff. Um, take it up with your town governance and really discuss this. Don't go out and commit crimes and bash things and commit, you know, acts that are now being called terrorism in Massachusetts. Towns like crazy are passing bans against us. So I like the idea and the safety and the evidence that comes from these cameras. I really do. Uh, I have a friend who got involved in a car accident in an intersection, and guess what? The person lied and said that they ran a red light and it was right there on the traffic cam, and they played that back and. Yeah, but I really hate the fact that. That it is so easy for people to abuse this stuff. I mean, when the mayor comes into the station house and says, I want you to tell me every time that person is out driving and where they went, what do you say? No. I mean, I would. And then I would get eased out the door, and my consulting contract would be canceled, and I'd have to get another job. But not everybody has that luxury. And not everybody would say no anyway. So a lot of places are just banning these damn things. Uh, I think I'm on board with that. Don't abuse this tech, though. Zero Trust is clearly the future, as, uh, threats get faster, quieter, and harder to detect. But implementing it shouldn't disrupt the business. Threat locker enforces default deny at execution in a way that remains enterprise ready, scalable, and operationally clean. Unknown software is stopped cold. Trusted apps stay contained, and drift is locked down across the environment. It's Zero Trust that works in real enterprises and prepares you for the threats ahead. See why CISOs are adopting it@securityweekly.com threatlocker and now, live from Caesar's palace in Las Vegas, where tomorrow night he'll be fighting for the heavyweight title of the world. It is Josh Barpet. Hi, Josh.

Speaker B: I. I promise you I'm not going to be fighting. And I'm not at Caesar's Palace. I'm in Tuscany Suites.

Speaker A: I wasn't going to disclose your location. I know you have all these groupies and fans chasing you down.

Speaker B: Uh, no, it's no big deal. I'm at Tuscany Suites. Where? B sides. Las Vegas was early in the week, and it was wonderful. It was a great conference. They had some great talks. Josh Corman spoke. Uh, they ran the cognitive security conference here yesterday and today. Uh, I did not. I made it to Blackhead a little bit. Uh, Mandalay Bay it is. It was a zoo go. If you tried. If you stayed in Mandalay Bay, God help you, because getting on the elevator was a 20 minute line. Not joking.

Speaker A: It's always a suit.

Speaker B: Oh, my God. And I didn't get to go. No, definitely going, um, on. And, uh, I got my badge. Uh, the badges are very, very cool. You've probably seen them online. But they're very, very neat. This is a removable module that afterwards will be a, uh, password manager and FIDO compliant module. Uh, and I saw Dallas from Hak 5 in the elevator last night, and he is the sweetest. Uh, he gave me one of the pineapple pagers. Uh, there you go. And they're now coming clear. So, like, there's also.

Speaker A: Josh always gets all the best swag. People, uh, like him.

Speaker B: I don't know why I'm not a nice.

Speaker A: The last time I saw somebody, one of those cons, he said, here, you want something, and handed me a bag. It was full of. It was full of dog crap. So it was.

Speaker B: I don't. I don't think so. But, um, some interesting things are happening. Uh, I managed to. I got invited to an roundtable yesterday at the Privacy Village. I'm speaking, by the way, Saturday at 5 o' clock on the Crazy Josh. Uh, I'm speaking on the Raise act, which is the, uh, AI Management act for the New York DFS Department of Financial Services and where they left some things out. So, anyway, uh, but my story is about. Actually came from this roundtable I went to about an open and interoperable web for agents. And by the way, just to be clear, black hat. If you went down on the black hat floor and all you saw was agent. That's correct. Agentic AI. Agentic AI. Sock AI. Sock AI guardrails for AI. AI security. AI implementation. Get rid of all your people with AI. Like, it's just, it's AI everything. This is the topic. There was nothing. Not AI there. Trust me. All right, but my story is about DNS. Huh. What they're trying to do is extend DNS so that you can query a DNS subtype for what agents are you running? So an agent, uh, an agentic AI would be seen as effectively a domain, and then sub agents would be subdomains. So you can have a globally unique id, a GUID or some kind of primary identifier for an identity, an agent. And then every. You can query for the capabilities. You can query for the, uh. What agents do you have? I argued that this is an attack surface. And they went, oh, we never thought about that. And they did a little bit. I'm not trying to impugn them. It was a lovely roundtable. But I put in the technical specification DNSAID where they're trying to make an agentic DNS. Now, in one sense, that's actually pretty smart. And it's a good idea to have a global unique identifier for every agent so that we can track what they do, who is accountable for them, what, uh, capabilities they have. On the other hand, it will just make it absolutely more true that in the end it's always DNS. Back to you, Doug.

Speaker A: Well, on that dark note. No, but you're right, I agree with that. It's still, it's still this DNS at the end of the day. Thank you, Josh. Glad you're having fun in Las Vegas. And I wish I was there, but I'm not. So, you know, it's just everybody will have to manage without me. I don't know how it'll go, but, you know, it's all right. Thank you. Uh, since we're just. I was in a very dystopian mood today and this last story was in the category of what could go wrong. Uh, LLMs are now building DNA and bacteriophages. Yeah. Now, a long time ago I got into coding genetic algorithms. So somebody gave me a book and it was called Genetic Algorithms. And I read it and I was like, this is so cool. So I made these structs and in c. And my structs were like DNA, right? I mean, I mean my struct had all these things in it. And then if two of my little things, my little structs get together and they love each other very much, they have a baby struct. Ooh. And the baby thing kind of gets a smattering of the structs of each parent and becomes a new struct. Oh, it was fun. And it was so weird and it was kind of scary. I mean, I let the SIM run on my. Like I had this old, uh, desktop and I was letting the SIM run on there because it was pretty slow. I just wanted to see what would happen. And it was interesting to see recessives that I injected in the thing emerge. And I didn't make any rules like siblings couldn't date or anything like that. And a lot of my little code babies ended up with virtual hemophilia, you know, that kind of thing. I think some of the things I included in my struct were werewolf DNA, gluttony, D baggery, people who take their shoes off on airplanes, you know, that kind of stuff. And those were the genetic components of these things. And I put a whole bunch of very, you know, tame stuff like blue eyes and brown eyes and black hair and black eyes and you, uh, know, deafness and all kinds of stuff that I just put in this struct that could possibly appear with percentages that told how likely those things were. And you know, I basically ended up with that guy over there in seat 19D right now, hairy D bag, stinky feet, shoes off, eating a giant container of hard boiled eggs, kimchi and kippers. But now this same Idea applies to LLMs, right? I mean, so basically one of the things that this article was about is they started making bacteriophages. And, uh, bacteriophages is a bacteria that can attack essentially a virus. And I'm not saying this is a bad thing either. I'm just saying it's a scary thing. So the phage, which this one that they came up with is called phex1, uh, 174. I don't know what the name means, but it's based off of E. Coli. And they fed millions of DNA sequences from viruses or from bacteria that infect viruses, or bacteriophages, and they put all those into the LLM and then they started building one to target E. Coli. Uh, does that sound scary yet? I hope so. I mean, I'm all for science, but this is terrifying. Take an amoral AI. I mean, we don't actually have those really at this point. We just have statisticians, which are pretty amoral too. But tell it to solve the hunger problem on the planet Earth and it builds a bacteriophage to target humans and poof. Yeah, I mean, we've come a long way from my little genetic algorithm struct porn, but we still don't have the three laws of robotics. And you know, in every single one of those futures, I don't care which ones they are, I have some bad news. They had to go through dystopia before they got to Utopia. I mean, you want the Federation, but look at Star Trek. They had the Second Civil War, the Eugenics Wars, World War 3, the Xindi conflict, the Romulan War, and the war to get Riker to force Riker to get tested, for God's sake, before they got to anything even remotely stable. So I don't know. I'm thinking the future is more William Gibson all the time. Which is sad, but. I mean, we can hope for the Jetsons, but I fear we may more likely to get A Clockwork Orange, Soylent Green and Mad Max. So, you know, Philip K. Dick all rolled up into one. So anyway, enjoy Vegas if you're there. Thank you for watching. Thank you, Josh, for taking time out of your busy day in Las Vegas to talk to us. And I will see you next time on the Security Weekly News. Thanks for watching. If you want more content like this, head over to securityweekly.com subscribe. You'll find all our shows, the latest episodes, and everything happening across the Security Weekly network. See you next time.