WEBVTT - Legendary Hacker Matt Suiche on Cyberwar in the Age of AI 0:00:02.720 --> 0:00:14.000 Bloomberg Audio Studios, Podcasts, Radio News. 0:00:18.560 --> 0:00:21.520 Hello and welcome to another episode of the Odd Thoughts podcast. 0:00:21.680 --> 0:00:23.080 I'm Tracy Alloway. 0:00:22.760 --> 0:00:24.319 And I'm Jill. Why isn't thal Joe. 0:00:24.360 --> 0:00:29.120 You know I have some uh prepper tendencies. Yeah, slightly 0:00:29.200 --> 0:00:32.200 prepper tendencies, Prepper adjacent. 0:00:31.920 --> 0:00:35.199 I know you do, because my plan for when everything, 0:00:35.760 --> 0:00:37.959 when everything goes bad, is to bring my family over 0:00:38.000 --> 0:00:38.600 to your place. 0:00:38.840 --> 0:00:39.760 So I'm relying on you. 0:00:39.840 --> 0:00:42.360 Actually, that's fine. I actually figured that, and I've been 0:00:42.400 --> 0:00:44.760 building an extra store of supplies. 0:00:44.880 --> 0:00:46.640 I'm gonna like send you a whole list of things 0:00:46.760 --> 0:00:48.839 my kids like to eat and stuff like that, just 0:00:48.880 --> 0:00:51.800 so that we're already Yeah, okay. 0:00:52.000 --> 0:00:54.440 Well, one of the things I saw on a bunch 0:00:54.440 --> 0:00:57.760 of the prepper boards that I sometimes look at. I 0:00:57.760 --> 0:00:59.600 don't want people to think that I'm crazy about it, 0:00:59.600 --> 0:01:01.960 but I find it interesting. I find it interesting seeing 0:01:02.000 --> 0:01:05.800 how people's like insecurities manifest in physical stuff. But anyway, 0:01:06.480 --> 0:01:08.640 one of the things everyone was saying was you need 0:01:08.680 --> 0:01:11.839 to start taking cash out because of the situation in Iran, 0:01:11.920 --> 0:01:15.080 because we're all expecting a big cyber attack that's going 0:01:15.120 --> 0:01:19.280 to absolutely destroy the US financial infrastructure. 0:01:19.440 --> 0:01:20.880 By the way, if I told you my idea for 0:01:20.959 --> 0:01:24.360 a business, like I've looked at prepper meals, like prepared meals, 0:01:24.360 --> 0:01:26.120 and they all look terrible. They do like a slightly 0:01:26.200 --> 0:01:29.399 high end version for yuppies I think would be really good, 0:01:29.680 --> 0:01:31.080 like something they you know, like. 0:01:31.080 --> 0:01:34.320 Some nice I think it's a physical limitation on how 0:01:34.560 --> 0:01:37.200 good you can actually get, like dry food. 0:01:37.240 --> 0:01:39.400 Scientist can do a lot of things these days. Anyway, 0:01:40.240 --> 0:01:43.240 let's talk about the actual issue at hand. Yeah, well, 0:01:43.280 --> 0:01:44.240 are you taking cash out? 0:01:44.800 --> 0:01:47.760 I have it? I haven't yet. I'm relying on my 0:01:47.840 --> 0:01:50.320 store of gold and silver. That's right. But I think 0:01:50.360 --> 0:01:55.360 this raises a legitimate and actually very interesting topic, which 0:01:55.400 --> 0:02:02.120 is what do we know about Iran's cyber I guess facilities, skills, 0:02:02.200 --> 0:02:06.440 what could happen in those context? And then also everything 0:02:06.440 --> 0:02:08.720 that's going on with the world of AI, right like 0:02:09.080 --> 0:02:14.200 cyber securities, cyber hacking. It's changing really rapidly given this 0:02:14.280 --> 0:02:15.679 new technology. 0:02:15.200 --> 0:02:18.000 Totally, I mean also just within the context of the 0:02:18.040 --> 0:02:21.919 war itself, setting aside hypothetical doom scenarios. There's a really 0:02:22.000 --> 0:02:25.680 interesting report in the Financial Times about Israel having been 0:02:25.720 --> 0:02:28.720 able to hack into all of the traffic lights in Tehran, 0:02:29.400 --> 0:02:33.839 almost unbelievable and shocking, but there's already within the war 0:02:33.880 --> 0:02:36.560 itself or even over the last couple of years, there 0:02:36.600 --> 0:02:41.040 was the pager attack that Israel had executed, and so yeah, 0:02:41.080 --> 0:02:43.280 cyber is part of it. And the timing is wild 0:02:43.360 --> 0:02:47.359 here because speaking of AI, it was just on Friday. 0:02:47.560 --> 0:02:49.840 We're recording this March fifth. I'm not exactly sure the 0:02:49.919 --> 0:02:52.320 data is coming out, but a week ago basically there 0:02:52.360 --> 0:02:56.359 was the news the complete collapse of the Anthropic relationship 0:02:56.360 --> 0:02:58.440 with the Department of Defense or the Department of War. 0:02:58.960 --> 0:03:01.160 And so it's all in the mix right now, and 0:03:01.200 --> 0:03:04.240 how is AI actually going to change warfare? And one 0:03:04.240 --> 0:03:07.440 of the national security implications of AI and AI and hacking. 0:03:07.800 --> 0:03:09.880 There is a lot in this sort of mix that's 0:03:09.919 --> 0:03:10.760 all happening right now. 0:03:10.800 --> 0:03:13.000 Absolutely. The thing that really caught my eye was the 0:03:13.080 --> 0:03:16.160 story about a hacker using Claude to hack into like 0:03:16.160 --> 0:03:17.960 the Mexican government system. Did you see that? 0:03:18.040 --> 0:03:21.080 That was really interesting because it seemed like the hacker 0:03:21.200 --> 0:03:24.440 extracted a bunch of information from Claude itself. You know, 0:03:24.520 --> 0:03:28.120 I'm pretty sure you cannot go to claud code and say, like, 0:03:28.160 --> 0:03:31.160 I want to break into the Mexican government website helped me, 0:03:31.200 --> 0:03:32.080 like build this app. 0:03:32.120 --> 0:03:32.800 It won't do that. 0:03:32.960 --> 0:03:37.080 It's trained to avoid malicious uses. But people find a 0:03:37.080 --> 0:03:39.240 way to jail break down. People find a way to 0:03:39.400 --> 0:03:42.320 sort of extract information from the AI itself that has 0:03:42.320 --> 0:03:45.680 in its training and so forth. And there's been examples 0:03:45.760 --> 0:03:48.520 of leaks where you know, people upload data to the 0:03:48.560 --> 0:03:51.920 AI and somehow other people see it. Anyway, there's a 0:03:51.960 --> 0:03:53.440 lot here that we have to learn more about. 0:03:53.560 --> 0:03:55.080 We should talk about all of it, and we do, 0:03:55.160 --> 0:03:58.120 in fact have the perfect guest, someone who's been on 0:03:58.160 --> 0:04:00.480 the podcast before but it's been a while. We're going 0:04:00.520 --> 0:04:02.800 to be speaking with Matt Swish. He is the founder 0:04:02.800 --> 0:04:07.360 of ondb, which is a data infrastructure startup for agentiic AI. 0:04:07.600 --> 0:04:12.160 So honestly the perfect a legendary French hacker. I should 0:04:12.200 --> 0:04:14.560 have said that first. Matt, thank you so much for 0:04:14.680 --> 0:04:15.760 coming back on all lots. 0:04:16.080 --> 0:04:18.360 Thank you very much. It's been I think it's been 0:04:18.480 --> 0:04:19.200 worth four years. 0:04:19.320 --> 0:04:20.200 Yeah, I think it has been. 0:04:20.440 --> 0:04:22.320 The Last time we spoke to you, you were still 0:04:22.480 --> 0:04:25.440 in Dubai, and now you're coming to us from Sweden 0:04:25.480 --> 0:04:28.200 and a very Gustavian looking background over there. 0:04:28.279 --> 0:04:30.359 I think actually what we talked to is right after 0:04:30.640 --> 0:04:33.440 Russia's invasion of Ukraine. So I guess, yeah, wow, that 0:04:33.480 --> 0:04:34.520 has been almost every. 0:04:34.400 --> 0:04:35.880 Time there's a war, we call you Matt. 0:04:35.920 --> 0:04:40.320 But because war is so intermixed with cyber espionage, cybersecurity, hacking, 0:04:40.320 --> 0:04:42.280 and so forth, it's a natural natural time. 0:04:42.480 --> 0:04:44.760 So for the benefit of people who didn't listen to 0:04:44.800 --> 0:04:47.040 the episode four years ago, can you just give some 0:04:47.080 --> 0:04:50.400 context around who you actually are and your sort of 0:04:50.680 --> 0:04:54.680 history in the hacking community, including you know, shadow brokers 0:04:54.680 --> 0:04:56.560 and the Wannacrey era and all that stuff. 0:04:56.800 --> 0:04:59.960 So I've been in enterprise software for almost like twenty years, 0:05:00.040 --> 0:05:05.760 particularly cyber security, and my name appeared in a few 0:05:06.040 --> 0:05:09.000 of the different leaks because of various analysis that I've 0:05:09.160 --> 0:05:12.760 done of private information that was being leaked, but also 0:05:12.760 --> 0:05:15.520 a lot of attacks that happened that happened to target 0:05:15.520 --> 0:05:19.280 critical infrastructure over the last ten years. And last time 0:05:19.320 --> 0:05:21.719 we were on the podcast, one of the things we 0:05:21.760 --> 0:05:27.080 talked about is does cyber really matter once you enter 0:05:27.160 --> 0:05:30.599 into a kinetic war, which is exactly what's happening now. 0:05:31.360 --> 0:05:35.720 And the main takeaway was once you start using missiles, 0:05:36.040 --> 0:05:38.760 most of these cyber elements are not really relevant because 0:05:38.800 --> 0:05:43.039 you would use cyber mostly to gather information and intelligence 0:05:43.400 --> 0:05:47.800 to prepare an attack or to disorganize an enemy. You 0:05:47.880 --> 0:05:52.320 create confusion, but as we have seen now, you can 0:05:52.480 --> 0:05:55.080 use like drones that are like twenty thousand dollars and 0:05:55.720 --> 0:05:58.880 create more chaos that you would do with any sort 0:05:58.920 --> 0:05:59.760 of exploits. 0:06:00.200 --> 0:06:02.880 I like how you're introduced as a legendary hacker and 0:06:02.880 --> 0:06:05.480 then you're like, oh, I've been twenty years in enterprise software. 0:06:05.640 --> 0:06:07.680 I feel like this is like the uh, the wind 0:06:07.720 --> 0:06:08.320 of the premium. 0:06:08.400 --> 0:06:10.880 It's like, you know, the casual and then fancy a 0:06:10.960 --> 0:06:16.560 dress up hacker and the like twenty years in enterprise software. 0:06:16.600 --> 0:06:19.520 But this is a really interesting point that you made 0:06:19.839 --> 0:06:23.760 this idea between Okay, mostly it sounds like when people 0:06:23.800 --> 0:06:28.080 imagined cyber attacks, they imagine what Tracy talked about in 0:06:28.120 --> 0:06:31.360 the beginning. Suddenly the entire financial infrastructure, like I just 0:06:31.360 --> 0:06:33.400 come to a haul of people worried about or there's 0:06:33.400 --> 0:06:36.720 gonna be a blackout, et cetera. But in reality, or 0:06:36.760 --> 0:06:39.560 what we've seen so far, by and large, is that 0:06:40.320 --> 0:06:43.239 cyber in the context of war is still much more 0:06:43.400 --> 0:06:47.839 about data collection, espionage, and so forth, rather than these 0:06:47.880 --> 0:06:49.680 more like you know, the types of things you might 0:06:49.720 --> 0:06:50.479 see in a movie. 0:06:50.839 --> 0:06:53.760 Yeah, exactly. I mean over the last like ten fifteen years, 0:06:53.800 --> 0:06:56.839 we've seen some like attacks like cyber attacks against a 0:06:56.839 --> 0:07:02.520 critical infrastructure you run talking like Aramco around like twenty twelve. 0:07:03.040 --> 0:07:05.520 They were just mostly like using what we call a 0:07:05.520 --> 0:07:08.800 wiper that was like a malware that was erasing the 0:07:08.800 --> 0:07:11.680 hard drive of most of the machines. And then we 0:07:11.760 --> 0:07:16.080 obviously have the case of Stocksnet few years before, where 0:07:16.360 --> 0:07:19.400 it was a joint Israeli US operation against some of 0:07:19.440 --> 0:07:23.120 the nuclear centrals in Iran where some of the PLCs 0:07:23.160 --> 0:07:26.720 were targeted. But what we have seen over the weekend 0:07:27.640 --> 0:07:30.480 is some of the drones happened to target some of 0:07:30.520 --> 0:07:35.360 the Amazon data centers and that created so much instability 0:07:35.720 --> 0:07:38.480 because multiple of the zones have been down and I 0:07:38.520 --> 0:07:41.400 think two out of three and the third one is 0:07:41.440 --> 0:07:47.000 still recovering days after. Because most of companies, either private 0:07:47.000 --> 0:07:50.400 companies or public companies not relying on the cloud, which 0:07:50.440 --> 0:07:53.800 is something that was not really the case before, and 0:07:54.280 --> 0:07:59.080 once you have some softw of centralization in terms of dependence, 0:08:00.080 --> 0:08:05.160 so become an easy target. And most of government's AI companies, 0:08:05.600 --> 0:08:10.239 cloud companies do not really have twenty thousand dollars drones 0:08:10.320 --> 0:08:13.720 in their threat models, which is like something that's pretty new, 0:08:13.760 --> 0:08:17.600 but also confirms that Kinneti quals and have more impact. 0:08:17.920 --> 0:08:21.120 So I take the point about cyber being perhaps more 0:08:21.200 --> 0:08:24.520 useful before a war when it comes to info gathering 0:08:24.600 --> 0:08:28.600 and things like that. But we have seen some deployment 0:08:28.680 --> 0:08:31.120 of cyber attacks in the past week or so. So 0:08:31.160 --> 0:08:34.480 we know Israel is attacking some cyber infrastructure in Iran, 0:08:34.600 --> 0:08:38.080 and we know that Iran has perhaps attempted some things, 0:08:38.240 --> 0:08:41.280 maybe not as successfully, but walk us through what we've 0:08:41.320 --> 0:08:42.520 actually seen so far. 0:08:43.160 --> 0:08:46.000 So so far, we have seen an Israeli operation where 0:08:46.080 --> 0:08:49.360 one of the prayer app has been hijacked and so 0:08:49.520 --> 0:08:53.160 message was sent to the users, so it's more like 0:08:53.360 --> 0:08:58.680 to create confusion within people. Also the traffic Light operation 0:08:59.000 --> 0:09:02.800 to understand the position of some of the targets, but 0:09:02.920 --> 0:09:07.400 it's more used for like recaoning a In terms of destruction, 0:09:08.000 --> 0:09:12.120 we didn't see anything significant. Even the government itself of 0:09:12.160 --> 0:09:15.840 Iran shut down most of the Internet for a lot 0:09:15.840 --> 0:09:17.920 of the users, and a lot of what we see 0:09:17.960 --> 0:09:21.760 on social media is the usual disinformation and misinformation campaign, 0:09:22.400 --> 0:09:24.720 especially now with AI. There is so much AI slob 0:09:25.200 --> 0:09:29.400 with like the videos, the text, the boats. That's becoming 0:09:29.440 --> 0:09:32.480 pretty common now, even when there is no war, So 0:09:32.679 --> 0:09:35.480 it's not really like really impactful. So it's more like 0:09:35.520 --> 0:09:39.800 to create confusion than being actually destructive. And now we're 0:09:39.840 --> 0:09:44.480 differently entering in a stage where it's been extremely destructive. 0:09:44.640 --> 0:09:47.200 And I cannot remember the last time we've seen so 0:09:47.240 --> 0:09:53.360 many countries being targeted, which is pretty like a first 0:09:53.440 --> 0:09:55.559 I would say, in terms of like a war climate. 0:09:56.040 --> 0:09:58.240 Can you talk about You know, people stare at their 0:09:58.280 --> 0:10:01.839 screens all day, and they fool themselves into thinking that 0:10:01.880 --> 0:10:05.480 they're quote monitoring the situation, et cetera, but mostly. 0:10:05.120 --> 0:10:06.080 The projections here. 0:10:07.120 --> 0:10:10.360 I don't delude myself, no, I like, actually like I 0:10:10.520 --> 0:10:12.280 sort of look at my screen and I know that 0:10:12.320 --> 0:10:18.120 I'm being inundated with contextless garbage and slop and propaganda 0:10:18.320 --> 0:10:22.479 and so forth. I'm curious how you monitor the situation actually, 0:10:22.640 --> 0:10:25.400 as someone who takes these topics seriously and doesn't just 0:10:25.440 --> 0:10:28.440 sort of become an overnight expert, you know, the day 0:10:28.520 --> 0:10:33.160 after bombings begin, Like how do you pay attention to 0:10:33.240 --> 0:10:36.079 what matters? How do you actually know what's real and 0:10:36.120 --> 0:10:38.960 so forth, and avoid just sort of the delusion of 0:10:39.040 --> 0:10:41.880 staring at the screen and engaging with slop. 0:10:42.280 --> 0:10:44.800 It's a good question because there's so much of it, 0:10:44.880 --> 0:10:48.560 so I think the default reaction is to ignore most 0:10:48.600 --> 0:10:52.640 of it, Yeah, unless it becomes really significant. In this case, 0:10:52.960 --> 0:10:55.640 I think it comes down to looking at the actual damage. 0:10:56.160 --> 0:10:59.400 Many people from the military world, but also the intelligence 0:10:59.400 --> 0:11:04.959 community has been underestimating Iran capabilities exactly like people used 0:11:05.000 --> 0:11:07.600 to do with North Korea, and now North koreas some 0:11:07.640 --> 0:11:09.800 of the best hackers in the world when we see 0:11:09.800 --> 0:11:14.319 them like targeting financial institutions, whereas before they would not 0:11:14.480 --> 0:11:18.120 do like much. So there is definitely like internal capabilities 0:11:18.200 --> 0:11:21.880 that are available, but there's so much noise now, Like 0:11:21.920 --> 0:11:24.000 you say, a lot of people are monitoring the situation 0:11:24.880 --> 0:11:29.199 giving that quote un quote like overnight expert opinions, and 0:11:29.320 --> 0:11:33.760 that's becoming a lot of noise. But I would say 0:11:33.920 --> 0:11:37.199 that in this particular case, we have heard of the 0:11:37.240 --> 0:11:42.800 imminent threat of Iran for around forty years, and that's 0:11:42.840 --> 0:11:46.520 also not really a new situation, so most of people 0:11:46.559 --> 0:11:51.640 would have context around it. And even for the attack 0:11:51.760 --> 0:11:54.760 that happened last week, and many people were expecting them 0:11:55.200 --> 0:12:09.800 four weeks, especially as the continuation of what happened last summer. 0:12:14.760 --> 0:12:16.760 Can you actually talk a little bit more about the 0:12:16.840 --> 0:12:20.599 data center attack? And because that's not cider really, I 0:12:20.600 --> 0:12:23.760 mean that's that's just a physical kinetic warfare against the 0:12:23.880 --> 0:12:29.200 data center. I was surprised how disruptive was that. I 0:12:29.280 --> 0:12:32.520 sort of would assume that cloud service providers that it's 0:12:32.559 --> 0:12:35.719 fairly liquid. Okay, one goes down, but you know, it 0:12:35.760 --> 0:12:38.280 can just be the same software, it can be run 0:12:38.360 --> 0:12:41.400 from numerous other clouds. But I saw that there were disruptions. 0:12:41.440 --> 0:12:44.280 I saw Fortnite tweeting about the fact that some of 0:12:44.280 --> 0:12:48.480 their gameplay was impaired due to the attack on data centers. 0:12:48.800 --> 0:12:51.920 How disruptive have those attacks been? Because this is of 0:12:51.960 --> 0:12:53.480 course a very. 0:12:53.440 --> 0:12:56.079 You know, that's whereinetic meatia. 0:12:56.160 --> 0:12:58.600 Yeah, yeah, and there's a lot you know in the future, 0:12:58.720 --> 0:13:01.920 like thinking about hardening these data centers and as you say, 0:13:02.040 --> 0:13:05.080 like making them they're going to be increasingly targets for war. 0:13:05.360 --> 0:13:06.439 Like how disruptive was that? 0:13:07.280 --> 0:13:11.360 Very good question? So I think one of the main 0:13:11.440 --> 0:13:15.680 takeaway is that it has been extremely successful. So, like 0:13:15.760 --> 0:13:18.000 we said before, like a Shahid drone is around twenty 0:13:18.040 --> 0:13:20.800 thousand dollars and they managed to shut down two of 0:13:20.840 --> 0:13:23.679 the zones of Amazon. Actually, even if you look at 0:13:23.720 --> 0:13:27.280 the official report from Amazon for like thirty six hours, 0:13:27.440 --> 0:13:30.000 they were just saying, oh, some objects struck the data 0:13:30.000 --> 0:13:34.520 centers before they actually expletely said they were drawn strikes. 0:13:34.840 --> 0:13:37.200 So a lot of services that I've been using them 0:13:37.320 --> 0:13:41.040 have been targeted, so from like local applications from two 0:13:41.080 --> 0:13:45.920 banks because in a data center you are taking care 0:13:46.000 --> 0:13:50.480 of multiple different services, and even Versail had to reroute 0:13:51.280 --> 0:13:56.359 their data to Bombay and to exclude middle List as deployment. 0:13:56.960 --> 0:14:00.280 So even if you take the cost of most of 0:14:00.320 --> 0:14:03.200 like zero, the exploits that can go up to like 0:14:04.000 --> 0:14:10.000 multimillion dollar attacks. If you are really aiming at districting things, 0:14:10.720 --> 0:14:17.040 the cost reward of using such an attack is really efficient. 0:14:17.120 --> 0:14:20.440 So you really enter into some sort of asymmetric conflict 0:14:20.720 --> 0:14:24.640 where you can just spend like some really old material 0:14:24.800 --> 0:14:27.240 and have way more impact than someone who's going to 0:14:27.240 --> 0:14:30.600 be like cutting edge and just trying to impress with 0:14:31.080 --> 0:14:33.040 like capabilities. Because at the end of the day, it 0:14:33.040 --> 0:14:34.320 does not really matter. 0:14:34.680 --> 0:14:38.120 How do governments actually build up their cyber capabilities nowadays, 0:14:38.160 --> 0:14:40.920 Because I have this image in my head of maybe 0:14:40.960 --> 0:14:44.000 ten or twenty years ago, you know, they would recruit 0:14:44.240 --> 0:14:47.680 like a twenty year old such as yourself at the time, 0:14:48.320 --> 0:14:50.760 and they would be working in a dark room, that 0:14:50.800 --> 0:14:53.640 sort of thing, but then drinking red bull. Drinking red bull, 0:14:53.680 --> 0:14:57.080 that's right. But then you know, we had the boom 0:14:57.240 --> 0:15:00.640 in Silicon Valley and so you had competition from private companies. 0:15:01.120 --> 0:15:04.240 Now we have the boom in AI and again even 0:15:04.320 --> 0:15:07.200 more competition from private companies. And at the same time, 0:15:07.720 --> 0:15:10.680 governments seem to be I guess, seeding some of their 0:15:10.720 --> 0:15:15.880 own skill set to potentially private companies like Anthropic and 0:15:16.080 --> 0:15:19.840 chat GPT and some others. Walk us through how I 0:15:19.880 --> 0:15:24.440 guess the development of governmental cyber capacity has actually shifted. 0:15:25.360 --> 0:15:28.320 I mean something that didn't really change over the last 0:15:28.800 --> 0:15:31.800 years in terms of capabilities, Like I guess we all 0:15:31.840 --> 0:15:36.520 remember like these snodnnicks in twenty thirteen when we started 0:15:36.560 --> 0:15:41.120 to see more about the inside of capabilities from a government, 0:15:41.880 --> 0:15:48.520 including like domestic my surveillance, global surveillance, exploitation capabilities, And 0:15:48.640 --> 0:15:52.680 since then every other year we have seen an history 0:15:52.880 --> 0:15:57.560 of data being leaked that belongs to the government. So 0:15:57.920 --> 0:16:00.400 in a way, things have been changing a lot, but 0:16:00.640 --> 0:16:04.800 not really much. Like most recently, there was a contractor 0:16:04.920 --> 0:16:10.560 from n Free Iris that was sentenced to eighty seven 0:16:10.600 --> 0:16:15.760 months sentence because they happened to sell zero the exploits 0:16:15.840 --> 0:16:22.560 to a Russian broker, and that's actual expert that belonged 0:16:22.560 --> 0:16:25.200 to the government because there was some sort of integrator. 0:16:25.720 --> 0:16:28.800 So we see like nation states or like governments like 0:16:28.840 --> 0:16:31.720 the US investing like an almost amount of money into 0:16:31.800 --> 0:16:35.880 offensive capabilities, but they also keep being burned by insiders. 0:16:36.680 --> 0:16:40.240 A lot of those capabilities are also as strong as the 0:16:40.280 --> 0:16:41.960 internal coercion. 0:16:42.240 --> 0:16:44.200 But I guess what I'm asking is, you know, if 0:16:44.240 --> 0:16:46.520 you're the Department of Defense or I guess now the 0:16:46.600 --> 0:16:50.480 Department of War, and you're thinking about developing in house 0:16:50.520 --> 0:16:55.400 capabilities versus partnering with a company like Anthropic, and we 0:16:55.400 --> 0:16:57.600 should talk about all the drama that's going on there, 0:16:57.760 --> 0:17:01.640 how are you balancing those decisions nowadays versus say, ten 0:17:01.720 --> 0:17:02.640 or twenty years ago. 0:17:03.240 --> 0:17:05.720 Like my understanding is that now a lot of it 0:17:05.760 --> 0:17:10.080 is also like outsourced because they cannot really develop as 0:17:10.119 --> 0:17:15.600 many capabilities internally. So now like we have seen with 0:17:16.800 --> 0:17:21.440 Entropic that it had been used in the Maduro operation 0:17:22.200 --> 0:17:26.000 and then after that, like there was a per out 0:17:26.040 --> 0:17:30.960 from Entropic because they said it was violating their ethical 0:17:31.119 --> 0:17:36.320 like standard like policy, Yeah, standards. So I would say, now, 0:17:36.520 --> 0:17:40.840 something that's really changing like very fast is the incorporation 0:17:40.960 --> 0:17:45.199 of like AI into tho's decisions. But as we all know, 0:17:45.359 --> 0:17:48.960 like AI can also hallucinate, so even Darly or the 0:17:49.080 --> 0:17:51.720 CEO of Entropics that it's definitely not in a state 0:17:51.760 --> 0:17:54.760 where it can be used for like fully autonomous decisions 0:17:54.800 --> 0:17:57.480 like that. So I would say, like the AI element 0:17:57.680 --> 0:18:00.040 would be like the main difference even we start to 0:18:00.080 --> 0:18:04.600 see it now for like exploit development or vulnerability discovery, 0:18:05.119 --> 0:18:08.120 but it's still too early to kind of like give 0:18:08.160 --> 0:18:12.120 a definitive like opinion about it. But overall, I would 0:18:12.119 --> 0:18:13.120 say it's very similar. 0:18:13.560 --> 0:18:16.240 Well, talk to us about exploit development, because I know 0:18:16.320 --> 0:18:18.760 that you can't just go to cloud code and say, like, 0:18:18.840 --> 0:18:23.000 I'm working on zero day mailware attack, help me figure 0:18:23.040 --> 0:18:26.040 this out. But you know, I also know that there 0:18:26.040 --> 0:18:29.240 are some very talented people who pride themselves in being 0:18:29.320 --> 0:18:33.240 able to jail break AI and illicit outputs that the 0:18:33.480 --> 0:18:36.919 labs do not want their AIS to produce. So do 0:18:36.960 --> 0:18:39.040 you have a sense how just within the pure like 0:18:39.359 --> 0:18:44.160 hacker community, AI is being employed today for these purposes 0:18:44.280 --> 0:18:46.120 or what they're able to get out of these tools. 0:18:46.400 --> 0:18:49.120 It's a good question. So, like we started to see 0:18:49.119 --> 0:18:53.720 people leveraging like AI for like bug discovery, which actually 0:18:53.920 --> 0:18:56.680 is becoming like pretty good. I think even Entropy published 0:18:56.680 --> 0:19:00.840 an article explaining how cloud can be used for the 0:19:01.560 --> 0:19:05.560 discovery into smart contracts and how it found some bugs automatically. 0:19:05.600 --> 0:19:07.720 And I think even recently there really is something called 0:19:08.720 --> 0:19:15.560 cloud for security that was aiming at doing cod assessment. 0:19:16.119 --> 0:19:20.080 But now we're entering in is like interesting paradigm shift 0:19:20.119 --> 0:19:24.439 where the cost of software is going towards zero. So 0:19:24.680 --> 0:19:27.400 if you're a company and if your cost of building 0:19:27.480 --> 0:19:31.480 software is becoming less and less, it's also hard to 0:19:31.520 --> 0:19:35.119 convince people that auditing software for security reasons is going 0:19:35.200 --> 0:19:38.000 to be more expensive than developing the actual software. So 0:19:38.040 --> 0:19:40.679 I think that's one of those shifts we're going to see. 0:19:41.000 --> 0:19:45.000 But when can you explain positive lives Mark, what did 0:19:45.040 --> 0:19:47.200 you mean by that? It's going to be hard to convent. 0:19:47.600 --> 0:19:52.479 If you can have like to allocate budgets for building 0:19:52.520 --> 0:19:55.040 a product, So you have most of the budget that's 0:19:55.119 --> 0:19:58.320 usually allocated for like your software engineers to build the software, 0:19:58.800 --> 0:20:02.840 and then you do some code review afterwards to make 0:20:02.840 --> 0:20:06.680 sure there's no vulnerabilities before it gets released to the public, 0:20:07.240 --> 0:20:11.320 but securities risk is usually like pretty high. You cannot 0:20:11.440 --> 0:20:14.480 just rely on like EI tools, at least not at 0:20:14.480 --> 0:20:17.040 the moment. Maybe in a year from now it's going 0:20:17.080 --> 0:20:20.879 to be possible. So it's going to be like really 0:20:20.880 --> 0:20:23.359 interesting to see how it's going to like do like 0:20:23.440 --> 0:20:27.680 a market shift because now with cloud code, and as 0:20:27.760 --> 0:20:31.600 a famous Vibe coder, Joe, I'm sure you know that 0:20:31.680 --> 0:20:35.640 the cost of building software is approaching like zero if 0:20:35.640 --> 0:20:37.400 you just look on the timeline. 0:20:37.600 --> 0:20:39.760 Are you actually on this note, are you a believer 0:20:39.920 --> 0:20:43.240 in the SaaS apocalypse? Because obviously there's the argument that, well, 0:20:43.280 --> 0:20:46.439 now everyone can just create their own software fairly easily 0:20:46.560 --> 0:20:49.040 using natural language. But on the other hand, if you 0:20:49.040 --> 0:20:52.760 are a big corporation or presumably a government you're going 0:20:52.840 --> 0:20:55.439 to want to have you're going to still want to 0:20:55.600 --> 0:20:59.439 buy software from an external provider given some of the 0:20:59.480 --> 0:21:04.399 security concerns, given that it might not necessarily make sense 0:21:04.440 --> 0:21:09.360 for various reasons, management reasons, perhaps recreate an entire software 0:21:09.400 --> 0:21:10.399 business in house. 0:21:10.920 --> 0:21:14.920 So I'm definitely biased on that. But as someone who 0:21:14.960 --> 0:21:17.280 thinks that the cost of software is going towards zero, 0:21:17.359 --> 0:21:22.040 and as someone who is like watching the software costs 0:21:22.040 --> 0:21:25.600 like collapsing. One of the things that we realize is 0:21:25.680 --> 0:21:28.560 that data is the only durabel asset in the AI economy. 0:21:28.640 --> 0:21:30.480 That's why we decided to work on the bid the 0:21:30.520 --> 0:21:34.760 current startup I'm working on, because like that's the only 0:21:34.800 --> 0:21:36.720 thing that's really going to have value long term if 0:21:36.760 --> 0:21:40.280 agents need something to transact or like to take the decision. 0:21:40.880 --> 0:21:43.800 Because even if you look in terms of like in 0:21:43.840 --> 0:21:48.720 any context, you know, if agents are designed to think autonomously, 0:21:49.240 --> 0:21:51.960 you need to have enough information to take those decisions 0:21:51.960 --> 0:21:56.040 whenever you're going to have your reasoning loops. So software itself, 0:21:56.119 --> 0:22:00.240 if you just build it, is pretty static, whereas like 0:22:00.800 --> 0:22:04.560 the urgentic like feedback loops are more dynamic. But what 0:22:04.680 --> 0:22:09.440 changes the contexts that take decision on. So definitely, like 0:22:09.720 --> 0:22:14.280 Salz business are going to have our time because if anyone, 0:22:14.480 --> 0:22:19.080 including the Shopify CEO, can just rewrite an MRI software 0:22:19.600 --> 0:22:23.480 in one afternoon just to like look at his back MRIs, 0:22:23.840 --> 0:22:26.919 you can imagine like how destructive it's going to be, 0:22:27.000 --> 0:22:29.040 like by the end of this year. I think the 0:22:29.080 --> 0:22:33.159 only thing we haven't seen yet is like enterprise AI agent. 0:22:33.720 --> 0:22:37.240 So so far I would say, like since Christmas, people 0:22:37.320 --> 0:22:39.800 are most still playing around trying to find like a 0:22:39.840 --> 0:22:44.280 proper use case. We see a lot of like consumer 0:22:44.440 --> 0:22:48.040 like AI agent like open Cloth that really met like 0:22:48.160 --> 0:22:52.879 agents more mainstream, but we really haven't seen yet enterprise 0:22:53.080 --> 0:22:56.280 AI agent. So as everyone is kind of scared of 0:22:56.320 --> 0:23:00.200 being replaced for their jobs, we haven't really seen in 0:23:00.240 --> 0:23:04.520 like actual like AI agents replacing entire departments or full 0:23:04.560 --> 0:23:07.800 on like employees. So we have seen some disruption around 0:23:07.840 --> 0:23:12.120 like software engineering, mostly to make like software engineering more efficient, 0:23:12.280 --> 0:23:16.720 especially in terms of like development with shorter timelines, but 0:23:16.840 --> 0:23:20.600 we haven't seen yet like proper like enterprise agents. 0:23:21.040 --> 0:23:22.600 How do you define an AI agent? 0:23:22.960 --> 0:23:25.439