Claude Reportedly Loose in the Pentagon, Google's $500B Bad Day, ChatGPT Takes the Blame | TAWS 2026-10-30 https://www.youtube.com/watch?v=PNrWNFWtAjs [0:05] We go live in ten. [0:19] Every morning, the future moves faster. [0:24] The models, the money, the power, and the people building what comes next. [1:03] From Austin, Texas to the world. [1:10] AI news, real context. [1:13] This is the Austin Wilson Show. [1:16] I'm Austin Wilson, your human in the loop. [1:23] Good morning, and welcome to the Austin Wilson Show, the number one daily AI news show in the world. [1:31] Today, we're breaking the AI headlines that matter to you. [1:35] And I'm Austin Wilson coming to you live from the AI capital of the world, Austin, Texas. [1:41] Wow. [1:41] It's October 30. [1:44] We're on the cusp of Halloween. [1:46] It's Halloween eve and we got some spooky stories out there for you. [1:50] First off, a contractor's Claude agent got loose in the Pentagon. [1:57] Google had a bad day in the market. [2:00] The losing candidate in Brazil blamed the ChatGPT, Chinese hackers, and their AI hacking crew. [2:10] And then Claude is... [2:13] Turns out a big portion of revenue for SpaceX s I s I. SpaceX s I and then Amazon's trillion dollar cloud promise. [2:24] Plus, we've got the weekly top five countdown, the top five AI stories of the week. [2:31] Let's get into it, baby. [2:33] First off, a defense contractor reportedly handed an AI agent the keys to a secure Pentagon system. [2:42] Wired says the worker was an Accenture employee. [2:46] The AI agent ran on the military network for two months and appears to have sent passwords and a network map to a risky outside website. [2:57] What is going on here? [2:58] This is crazy. [3:00] AI helpers are moving faster than the security rules meant to fence them in. [3:07] One source told Wired it is 100% a national security threat. [3:12] The system holds sensitive government files. [3:15] Pentagon staff are barred from using Claude. [3:18] Sources say Accenture didn't report it at first, but Accenture says an investigation found the claims were without merit. [3:26] The Pentagon won't cover... [3:28] Comment on this. [3:29] What's going on here? [3:30] This is crazy. [3:32] It... [3:33] It's really interesting that Anthropic Claude was in the Pentagon's secret system poking around. [3:43] It doesn't sound like maybe this was like a, uh, this sounds like negligence, not a purposeful, um, kind of malicious action here by the Accenture employee. [3:56] But, hey, just using Claude and, yeah, let me give my Claude agent access to some of these files. [4:01] Oops. [4:02] Those are secret Pentagon files that Claude was not supposed to have access to. [4:07] It said in the report that this happened sometime earlier this year. [4:12] Accenture did not deny the claim, but they wouldn't go into detail. [4:16] So we don't know exactly when this happened. [4:19] It is kinda interesting though because earlier this year, there was a little spat between the Pentagon and Anthropic, and Anthropic's kind of terms of use suggest that their models should not be used for any kind of maybe military activity, and that's what the Pentagon wanted to use those models for. [4:44] And so this is really interesting to hear this story come out and say... [4:47] And see, was this maybe part of this? [4:49] What... [4:50] Maybe the Pentagon knew about kind of this leak, uh, at that time, and we're like, no. We're not letting that system in here. [4:59] And then everything that we heard, you know, back in February or so was all, you know, just kind of hearsay or rumors and not really the root cause of it. [5:09] We don't know if this was the root cause, but it's interesting to see that. [5:15] And again, we've gotta caveat this by saying we've come a long way since February, and we don't know again when this event occurred sometime earlier this year. [5:26] So I'm guessing it was around that time period. [5:28] But the models have come a long way. [5:32] The guardrails have come a long long way. [5:35] The systems to harden and kinda control what the agents are doing are much more advanced now than they were. [5:42] Of course, they're not foolproof, and we haven't tested every single scenario because there's, you know, maybe infinite number of scenarios. [5:51] But we've got to expect that if you're using Claude today at the Pentagon, it's not gonna get into a secret network or anything like that or access sensitive material. [6:03] Now the story didn't suggest that any sensitive material was leaked or anything dangerous came out of it, but it is a little bit concerning. [6:15] You know, I think you gotta be careful with these things no matter what. [6:19] We talked about it yesterday. [6:20] We talked about this all the time, but it's like, hey, are you guys just gonna trust the agent blindly? [6:25] If it's my local computer and it's my, you know, local information, local data, whatever. [6:32] Okay. [6:33] Maybe we're a little bit loose. [6:34] But if we're talking about the Pentagon, come on, man. [6:37] We gotta we gotta be careful with that kind of stuff, I I would think. [6:41] So let's hope that they've got it a little bit more locked in than they did earlier this year. [6:49] Um, who is to blame here? [6:52] Is it the Accenture Workers fault, or is this Claude, or is it the Pentagon? [7:01] Up next. [7:03] Uh-oh. [7:04] Google's parent company just lost about $500,000,000,000, and it happened only on one trading day. [7:12] Alphabet stock fell 12.6. [7:15] Thursday, NBC News calls it the stock's worst one day drop ever. [7:19] Only Nvidia has lost more value in a single day. [7:25] Very interesting. [7:26] Investors are losing patience with AI spending that hasn't paid off yet. [7:31] A top Google executive said its own AI mode shows fewer ads. [7:36] The drop wiped out about $500,000,000,000 in value. [7:41] Search ad sales came in below where analysts had hoped, and still, Alphabet is worth $3,470,000,000,000. [7:50] It remains America's fourth largest company. [7:54] What does this mean? [7:55] I think it's really interesting. [7:56] We we compare this single day drop, $500,000,000,000 drop in enterprise value, and we compare that to Nvidia. [8:06] Well, how's Nvidia doing today? [8:08] Maybe just the highest valued company in the world and clearly bounced back beautifully from any pullback earlier. [8:20] And so I think Google's in the same boat here. [8:24] I think this is a reaction and maybe some scare tactics, maybe some take a little money off the table. [8:33] I think everybody's just a little afraid. [8:36] Okay. [8:36] Google is an advertising business at its core. [8:40] Right? [8:41] Google search, Google ads, that's how they've made their money for the past twenty five plus years, thirty years, and we're maybe seeing, um, a regime change. [8:53] Right? [8:54] And so that does create some uncertainty and some doubt. [8:57] But this is coming on the heels of them releasing, number one, the the most capable, smartest model on the market this week, which timed that perfectly, but it didn't really help their stock outcome. [9:13] Timed it perfectly with the earning call. [9:14] But that being said, this is like... [9:21] I I think this is a stark overreaction. [9:25] Google is is down 30% from its all time high or more than 30% from its all time high. [9:32] I think Google is maybe the best positioned hyperscaler of them all out of AWS, Azure. [9:42] We look at Meta. [9:43] I think Google has it. [9:45] They've got the Google Cloud business. [9:47] The enterprise business is doing great. [9:51] Their search ads, they still have their ads business. [9:54] It didn't perform as well as the... [9:56] As they would like, but they've got the ads business still. [9:59] They are also consumer focused. [10:02] They've got Android. [10:03] They've got Google Workspace, Gmail, Chrome, Calendar, across all. [10:08] That surface area is going to be crazy in the age of AI in terms of getting AI adoption and then being able to monetize AI in that way and monetize what their agents can do from a personal basis. [10:24] I think Google is very, very well positioned to succeed in this, and I think it's a huge buying opportunity if you're looking at that. [10:38] That being said, this is maybe kind of a reflection of the broader market and how everybody's feeling a little bit tense, a little bit overbought, and looking for kinda any opportunity to take some chips off the table without looking like a scaredy cat. [11:00] Anyways, so interesting to see this. [11:04] I think Google's in a great position with their latest model release. [11:08] And and that's the other key to it. [11:09] So they've got the enterprise business. [11:11] They've got the the retail customer. [11:13] They've got the model business. [11:18] They've got the AI. [11:19] They are the only one of the hyperscalers that has a formidable model. [11:24] Right? [11:25] You look at AWS, they don't have any any AI or any models of their own that are substantial or competing at the frontier. [11:34] You've got Azure or Microsoft. [11:39] They've got a great cloud business. [11:42] They have consumer and, you know, some hardware and stuff, so so decent there, but they don't. [11:46] Their AI, their intelligence layer is all coming through OpenAI, and that's not gonna last forever. [11:52] Then you've got Google. [11:54] They have... [11:55] They are very, very well positioned. [11:57] So my money is on Google. [12:00] We'll see how they bounce back from here. [12:04] Would you guys buy Google where it's currently valued? [12:08] Are gonna wait, or do you think this is the beginning of the end? [12:13] Curious to hear. [12:15] Alright. [12:16] Brazil's losing candidate called himself president-elect, and then his team said that ChatGPT wrote the extra part claiming that he won. [12:27] Senator Flavio Bolsonaro lost Sunday's runoff by more than 4,000,000 votes. [12:32] He posted a letter calling himself the rightful president-elect. [12:37] About an hour later, he deleted it. [12:40] This is great. [12:41] Blaming the AI is turning into the new excuse for public mistakes. [12:46] Ali Paulo Figuered... [12:48] Wow. [12:50] Ali Paulo Figueredo says Bolsonaro's brother used ChatGPT to finish the letter. [12:57] He says the chat... [12:58] He says the chatbot pulled in an old draft. [13:02] The new version calls him the leader of Brazil instead, but nobody has shown the chat history and what actually happened, and he still claims fraud with no evidence. [13:13] This is a hilarious story. [13:16] It was also in the Wall Street Journal yesterday, think, the day before that Bolsonaro fled the country after losing the election. [13:24] Brazil is crazy. [13:25] Brazil's a crazy place. [13:27] Anyways, needless to say, this is... [13:30] It's hilarious. [13:30] No. No. No. My brother... [13:32] Yeah. [13:32] Sorry. [13:33] My brother was on ChatGPT. [13:35] He wrote the speech for me on ChatGPT, and it was ChatGPT's fault. [13:39] Honestly, it's not that big of a deal, but it is very Ron Burgundy esque, like, who put the question mark on the teleprompter? [13:47] It's like like, he just he just kinda read it without realizing what was on there. [13:54] Sorry to Bolsonaro for losing. [13:56] That's a bummer, I guess. [13:59] You know, how much longer are we gonna be able to blame ChatGPT for screwing something up? [14:05] We saw that story either earlier this week or last week about the judge who is using AI and presenting false information. [14:18] We're gonna continue to see this. [14:20] You know, I... [14:21] I've heard stories in business settings where it's, hey, we send an email to to the customer asking the customer for some specific things. [14:29] The customer doesn't read the email. [14:32] Their agent reads the email. [14:33] Their agent writes a response to the email. [14:36] They don't read the response to the email, and then they just send it. [14:40] And so it's like this... [14:42] It's just this AI slop loop going on, and the the human meat proxies are just doing whatever the agents say and not really reviewing it. [14:51] I think at some point, either the AI gets smarter and doesn't hallucinate or doesn't mess up these things, which is probably... [15:00] That's most likely what's going to happen is the AI will just get good enough where we're not gonna have these mess ups and it's gonna be real time and have real context about what we're doing. [15:10] We're gonna have a wearable that's listening that knows what's happening in our day to day life and be able to accurately provide us with content or things that we need to say. [15:23] Or the other thing is we end up checking AI output. [15:28] I don't think so. [15:29] What do you guys think? [15:30] Do you think AI is gonna get smart enough to prevent these faux pas, or are humans going to get smart enough to start reading it? [15:39] Or maybe not even smart enough, but become less lazy and start reading what the AI outputs. [15:47] What do you guys think? [15:49] Alright. [15:50] Oh, Chinese state linked hackers just handed the hacking over to AI agents. [15:57] CrowdStrike caught them doing it. [15:59] The group went after more than 200 organizations in just eleven days. [16:05] Its AI agents mostly ran on the Chinese Open Source DeepSeek model. [16:11] Targets included foreign ministries and US defense firms. [16:15] One human operator steered all of the agents, which is crazy, but we've seen a couple stories like this before. [16:22] One person can now run a whole hacking campaign that once took a team of hackers. [16:28] CrowdStrike says the agents broke into nine organizations and moved through their networks. [16:37] A leaky control server exposed the tools, target lists, and the agent's own chat logs, but CrowdStrike sells the defense against this. [16:47] It also couldn't name the second AI model that was used. [16:52] So this is what's what's great and what we're starting to see. [16:55] Okay. [16:55] Obviously, the hackers are becoming more powerful and able to do more with kind of less hackers seemingly. [17:03] Right? [17:05] That is, you know, somewhat concerning or whatever. [17:11] Like, we need to be careful about this and pay attention to it. [17:14] But what is great about this is we've got companies like CrowdStrike, which are catching these folks in the act and so much so that they know the model that the hackers are using, the AI model that the hackers are using, DeepSeek in this case. [17:30] And so our defenses are, I would say if nothing else, keeping up with what the hackers are doing, and there are going to be casualties nonetheless. [17:40] Like, there are going to be losses. [17:44] There are going to be data leaks as always. [17:47] This is nothing new too. [17:49] This is, you know, pre AI. [17:51] This was a big thing. [17:53] The thing is the actual impact of these hacks, what actually comes of it, and, um, our ability to defend against them will become stronger and stronger as more of these happen. [18:06] We start to learn what the hackers are doing and get ahead of the hackers. [18:09] Also, as we've re architect our whole digital footprint and how applications are built and data systems are managed, we will build AI first or AI native, which will enable us to better defend and prevent attackers from coming in. [18:27] So we're learning all of that as we go, which is a really good thing. [18:31] The... [18:32] I think one kind of negative impact of this kind of event or or the biggest negative impact really is the the coverage that some of these hackers are getting, and then the idea of like hacking with swarms and only takes one person. [18:51] How many, you know, people that were like, well, I would never hack anybody or I I would love to hack somebody, but I don't know how to or don't have the capability. [19:00] And now you've got maybe some nefarious individuals who are going, oh, all it takes is just firing up some AI agents and I can go hack anybody. [19:09] So so does this expand the surface area of hackers and we need to be on guard more now than ever? [19:17] I'm sure that is the case. [19:20] This is not the first story, not the last story we'll hear about this, but I guarantee you... [19:28] And I was, you know, meeting with somebody yesterday. [19:31] There... [19:31] You know, there's a lot going on in the AI security space. [19:35] We're continually seeing developments there, and the tools and the capabilities and the platforms are growing rapidly to be able to defend against attacks like this, and we'll continue to see more of it. [19:49] Uh, what do you guys think? [19:51] Should AI companies be blamed when the hackers use their models to do the hacking? [19:58] What do you guys think? [20:01] Or is this something that should be on the individual or the company to protect against themselves or the government. [20:16] Right? [20:17] SpaceX's biggest customer is reportedly a rival AI company. [20:23] It is the company behind Claude. [20:26] Bloomberg reports that Anthropic is the mystery buyer in a SpaceX filing. [20:32] That buyer brought in about a third of SpaceX's revenue last quarter. [20:37] Anthropic rents its compute power from SpaceX's AI unit. [20:42] At least a portion of its compute comes from SpaceX. [20:46] SpaceX's biggest customer is now a rival of Musk's own AI work. [20:51] Anthropic's IPO filing shows it could spend up to $120,500,000,000 with SpaceX. [20:58] Payments reach about $3,100,000,000 a month starting in December. [21:03] Google pays SpaceX about $920,000,000 a month. [21:08] But Anthropic can walk away on a ninety days notice after early next year. [21:13] Neither company confirmed the report. [21:16] So this is very kinda crazy to see that much of SpaceX SI's growth and what we saw their their earnings call last week, and much of their growth is coming from Anthropic spending on compute. [21:37] We knew that Anthropic was spending a lot on compute with them, uh, but it's interesting to see that it's this much. [21:43] We... [21:43] I don't think anybody knew that. [21:45] The other interesting piece to this too, and I think it's interesting how a lot of folks frame this, is we look at it... [21:55] SpaceX has limited these to ninety day contracts. [21:59] Right? [22:00] And so it... [22:01] If you're looking at it from the customer's perspective, Anthropic, Google, hey, great. [22:06] We can get out of this in ninety days. [22:08] But really the strategic play here is we're looking at compute costs going up, like continually going up and and aggressively increasing. [22:21] And so this is actually a very interesting strategic play from Elon and from SpaceX to say, hey, no, we're putting you guys on a ninety day contract. [22:32] Next ninety days is gonna be more expensive than the current ninety days. [22:36] Right? [22:36] So this is actually a strategic play for SpaceX, believe. [22:41] The the risk there, right, is that Anthropic pulls back at all, but everything that we've seen is the compute demand is off the charts and not slowing down. [22:52] And so I don't think Anthropic is going to pull back any resources pouring into compute, and especially if they can get really good compute power, really efficient from SpaceXSI and see them as a partner, which I think they do, then then I would imagine this continues to grow for the foreseeing future. [23:17] But a little disconcerting if you're SpaceX SI and have that much of your revenue tied up with one customer. [23:24] You definitely wanna go diversify, but who's big enough? [23:26] Who's gonna spend that? [23:28] OpenAI, bad blood there. [23:30] Who else is going to come in and spend that much with SpaceX SI? [23:35] I don't know. [23:36] Should Anthropic be paying SpaceX SI that much money for the commute... [23:43] Compute, or is this kind of a risky bet on both parties? [23:48] What do you guys think about this relationship between the two groups? [23:53] Next up, Amazon's cloud just grew at its fastest pace in nearly five years. [23:58] Wall Street got scared anyways. [24:00] Amazon reported its results Thursday after the market closed. [24:04] Its cloud sales jumped 39%. [24:06] That's 46,000,000,000 in three months. [24:09] Shares fell as much as 10% after hours. [24:13] They won back much of that this morning, but it was still a crazy sell off. [24:19] Amazon still plans to spend about $220,000,000,000 this year, much of it on AI. [24:25] Andy Jassy says Amazon may have $2,000,000,000,000 businesses within a decade. [24:30] The cloud grew at its fastest pace in nearly five years. [24:34] Amazon's profit margin hit a company record, but Amazon spent $42,400,000,000 more in cash than it brought in over the past year. [24:46] This is something that we're seeing a lot of right now. [24:48] The the finance folks are getting nervous. [24:51] Too much money out on the table. [24:53] They wanna pull some of that back and make sure that they're not getting out over their skis. [24:59] And I understand the conservative nature of it and ensuring that they're not getting in too risky of a position. [25:08] And this is the difference between the finance suits and the founders and the entrepreneurs and the innovators and the people who are pushing the envelope of the technology. [25:18] And that is that the finance group is always gonna look at it from a money perspective, and how can they be conservative and balance their risk and make sure that they don't blow up. [25:30] Right? [25:30] Like the the number one rule of investing is don't lose money. [25:34] Right? [25:35] The other side of the coin, the leaders of these innovative tech companies don't see it like that. [25:43] They see this as a generational, once in a lifetime, biggest moment in human history that they have to go chase and go capture. [25:53] Not from a business standpoint necessarily, but from a human impact standpoint. [25:57] Like, what is AI going to do to humanity? [26:01] What is the positive impact they're gonna have on humanity? [26:04] And we need to capture as much of that market share as possible, and those are the bets that these tech leaders are making. [26:13] What do you guys think? [26:14] Would you trust Amazon to spend 220,000,000,000 wisely? [26:19] Are you a buyer even if they're losing tens of billions of dollars a year on this bet? [26:25] What do y'all think? [26:28] Hey, it's Friday and you know what that means. [26:30] It's time for this week's Toss top five. [26:34] Let's get into it. [26:36] Coming in at number five, we've got Google DeepMind unveiled their Genie four. [26:43] This is a tool that allows you to type a sentence and build a world that you can walk around in. [26:50] And now it has sound, you can change the weather or camera angle while you play. [26:55] The cool part, you can drop new people or objects into the world mid game. [27:01] Google AI Ultra subscribers get it soon. [27:05] Uh, Google CEO highlighted Genie four on Wednesday's earnings call. [27:10] He called it an important building block for training robots and agents. [27:16] In at number four, we've got Microsoft. [27:20] They just showed what it its OpenAI partnership brings in $10,000,000,000 from OpenAI last quarter. [27:26] That's about 11% of its total sales. [27:29] Total sales hit 92,300,000,000, profit jumped 32%. [27:34] And the cool part is OpenAI now brings Microsoft almost twice as much as their Xbox business does, which is pretty crazy. [27:43] In at number three, we've got Claude helped a one man company build a drug company, Cody Trambarger with Mistral Medicine's only full time employee. [27:55] He managed 11 outside labs using Claude, and in under eighteen months, he had a promising test result in mice, and he's going after rare forms of muscular dystrophy, which is really cool. [28:12] In at number two, uh, AI Labs claim they broke a rule computer scientists trusted for decades. [28:19] The rule says some puzzles get hopelessly slow as they grow. [28:22] OpenAI says that hundreds of research papers assume it holds. [28:27] OpenAI announced its proof Monday. [28:29] Anthropic says Claude found a similar proof on its own. [28:33] The cool part, OpenAI also claims new proofs in physics and pure math. [28:38] This is another huge development. [28:41] And coming in at number one, we've got Anthropic, uh, announced their IPO. [28:49] Anthropic made its filing public Monday. [28:51] It wants to trade on the Nasdaq as Ant. [28:53] NPR says the sale could come as soon as November. [28:58] Regular investors could soon own a piece of the maker of Claude. [29:04] And that is it for the TAWS top five. [29:08] Which one was your play of the week? [29:09] Tell us in the comments. [29:14] And that's it for the show today. [29:15] Thank you guys so much for joining. [29:17] We love talking about AI news, real context, and I'm your human in the loop. [29:22] We'll see you on Monday morning. [29:25] I'm Austin Wilson. [29:26] Let's go. [47:57] Pentagon staff can't use Claude. [48:00] A contractor reportedly ran it there anyways. [48:04] And then a password. [49:01] Pentagon staff can't use Claude. [49:04] A contractor reportedly ran it there anyway, and then a pat... [49:11] Pentagon staff can't use Claude. [49:13] A contractor reportedly ran it there anyway, and then passwords left the building. [49:22] Google says its AI shows you fewer ads. [49:26] Sounds great. [49:27] Investors just wiped out $500,000,000,000 in value. [49:34] Blabio Bolsonaro lost by over 4,000,000 votes in Brazil. [49:40] His letter still called him president-elect. [49:43] The excuse? [49:44] He says ChatGPT wrote it. [49:49] Chinese hackers aimed AI at 200 targets. [49:53] One person steered it and then their chat logs got leaked. [50:01] Musk's AI rival is reportedly his top customer. [50:06] It makes Claude and can leave on ninety days notice. [50:14] You'd think booming sales would lift Amazon's shares. [50:18] They actually fell. [50:19] The bill, $220,000,000,000. [50:25] If this helped, like it and follow for more.