October 22, 2026 2026-10-22 https://www.youtube.com/watch?v=cj4Vh4DluA4 [0:07] We go live in ten. [0:20] Every morning, the future moves faster. [0:26] The models, the money, the power, and the people building what comes next. [1:04] From Austin, Texas to the world. [1:11] AI news, real context. [1:14] This is the Austin Wilson show. [1:17] I'm Austin Wilson, your human in the loop. [1:25] Good morning and welcome to the Austin Wilson Show where we are breaking the AI headlines that matter to you today. [1:33] I'm Austin Wilson coming to you live from the AI capital of the world, Austin, Texas. [1:39] Today, we've got a big docket. [1:42] A lot of important things happening in the last twenty four hours. [1:45] First off, the AI that found flaws under most of the Internet. [1:49] Barclays Bank is cutting 5,000 jobs as AI moves in. [1:54] Meta's new AI stock and Zuckerberg's tax drama. [1:59] Anthropic's Claude duet agent, their personal agent is coming out imminently. [2:06] Senators grill Google over AI in classrooms and Anthropic's warning about AI training AI. [2:16] Let's get into it. [2:18] Up first, researchers using AI found two holes in the code behind most websites. [2:24] Governments pulled sites offline. [2:26] Hackers were already trying to infiltrate. [2:30] The flaws sit in PHP. [2:33] That code runs an estimated 70% of websites. [2:36] Teams working with OpenAI and Google's Wiz used AI to find the holes. [2:42] The fix came out Wednesday. [2:45] Amazon logged an attack attempt four days earlier. [2:49] This is big news coming out of the cyber world. [2:52] AI News finds bugs that hid for decade. [2:57] Excuse me. [2:57] AI now finds bugs that have been hidden for decades. [3:00] Attackers can use the same tools to, uh, exploit these weaknesses. [3:06] OpenAI called it one of the most impactful vulnerabilities in recent memory. [3:12] The NSA told site owners to patch now. [3:16] Sites that can't patch should go offline, it said. [3:19] But AI cuts both ways. [3:21] Amazon says hackers are now exploiting the flaws widely. [3:27] This is a big development in the past, uh, twenty four, forty eight hours. [3:32] We've seen several key, uh, tools and foundational technology to most web applications, uh, find and uncover weaknesses that are now being exploited, and they're constantly having to patch these vulnerabilities. [3:49] Uh, we saw this with PHP, Node, and several other technologies, uh, and and underlying tools that run our day to day, uh, Internet experiences. [4:02] What does this mean? [4:03] Well, number one, this is something that we've kind of been preparing for, and we've covered several stories that are very similar to this, and that the technology is moving so fast. [4:18] The intelligence is very, um, extensible. [4:23] So a lot of people can use this and exploit different weaknesses that previously nobody knew about. [4:32] And the way that they're doing this right is that when you have a swarm of AI agents that can go after these technologies, know, the the PHP and the Node, these underlying tools that run the Internet, if you can unleash a swarm of agents on it, you you you can find any vulnerabilities that might be in place, which it's a little bit scary. [4:58] Right? [4:59] That being said, we have powerful AI on the backside of it that is uncovering these weaknesses just as fast and patching them as the weaknesses are revealed. [5:13] And so this is a race, really. [5:16] We're seeing the good guys versus the bad guys. [5:19] Essentially, the good guys trying to find any weaknesses and fix those before the bad guys can get in. [5:25] The good news is, theoretically, the good guys have much more powerful technology from the frontier companies. [5:35] This is where OpenAI and Anthropic come in and can provide support to these businesses that may be in a little bit of a tight situation. [5:46] You know, we saw OpenCode a couple days ago. [5:49] We're seeing kinda more of these attempted attacks come out, and it's really important. [5:58] Right now is like crunch time for these AI companies really to figure out how do we harden all this and then share the information broadly so that everybody can protect themselves, that we can get really healthy, strong safety protocols in place to make sure that we are secure. [6:20] Now this isn't something that we just say, hey, unleash AI to solve AI. [6:26] Like, you have to be really thoughtful, and luckily, there's really, really smart people working on this, and I think we're in good hands. [6:34] That being said, it is, you know, a little bit concerning and doesn't go without, you know, alarm and making sure that you are protected, that your data is protected, and that you're paying attention to what is going on out there. [6:54] Some, you know, additional reports I saw come out this morning where some banks, uh, have seen third party agents acting on behalf of users, and they're not certain if these are personal agents like the MetaM user, Grok bot, or any of these other agents working on behalf of the the users and user directed, or if these are bad actors getting in there. [7:18] So we are in the early days of figuring all of this stuff out. [7:23] That being said, there's no need to freak out about this. [7:28] This is not... [7:29] I think this is to be expected. [7:31] Everybody has been expecting it. [7:33] The the fact is we need to be cautious. [7:37] Be cautious in what data you're releasing. [7:41] And and, you know, you think about the early days of the Internet and... [7:45] I mean, not even just the early days, but up till a few years ago when you have phishing emails and you have ways that people were trying to exploit individuals and, you know, for malintended purposes. [8:01] And we see similar stuff happen here. [8:03] This is just a new form factor. [8:06] So we gotta figure out how to fight it and win. [8:10] My money's on the good guys. [8:13] What's what's your opinion here? [8:16] Does AI make the Internet safer or more dangerous? [8:23] Moving on. [8:25] Barclays is cutting about 5,000 jobs. [8:28] At the same time, it's bringing more AI into the workplace. [8:33] About 1,000 roles are already gone. [8:36] The rest go by the end of this year. [8:38] That's about 5% of its global staff. [8:41] Most cuts hit support and operation jobs. [8:44] Software developers are among those being cut. [8:49] This is a big latest cut here. [8:52] It's not just Barclays. [8:53] HSBC and Standard Chartered have also flagged AI job cuts this year. [8:59] Barclays says years of tech spending let it run with fewer people. [9:04] Its profit still rose 12% last quarter. [9:07] The bank expects another £1,100,000,000 in savings next year. [9:12] But its CEO says AI is the icing on the cake. [9:16] It's interesting to see this come out of the financial sector where typically you think of banking as being maybe slower adopters to the technology. [9:30] But in this case, it seems that Barclays has found efficiencies with initial technology implementation and now seeing even further efficiencies and unfortunately leading to a layoff here. [9:46] But this is something that I think we'll continue to see more of. [9:49] Right? [9:50] Is these companies finding that they can make their employees more effective, the ones that know how to use the AI tools. [10:02] And what's interesting here is, and and what I'm seeing and hearing a lot of is the people that are able to 10 x themselves with AI tools are the ones that are keeping the job. [10:16] So it's super important right now to be learning and leveraging these tools as much as possible, not just for job safety, but your your life and livelihood is gonna be much richer because of the productivity that you'll find in using the tools, which I think is a really key point. [10:39] And if you're not learning these, go jump on YouTube and and go watch as many videos as you can about how people are using these tools. [10:47] I I was talking to somebody earlier this morning. [10:49] It's really interesting. [10:51] We look at the problems in our day to day life and typically, you know, either personal or or work wise. [10:59] And typically, you just solve the problems how you used to solve them. [11:04] Right? [11:04] You just did the work and you got it done and whatever it may be. [11:08] Now, it's really helpful and you should have a paradigm shift where you start thinking about instead of just addressing the problem as is, you think about how can you leverage AI to solve that problem, or how can you leverage technology to solve that problem. [11:27] And you'll be really surprised to find that there's a lot of solutions out there that weren't available before. [11:34] And that's personally how I've seen kind of the shift in my own life is, wow, what if I leverage the latest technology to solve some of these problems for myself? [11:45] What would that do to my business and to my personal life and something that I think would benefit you? [11:53] That being said, we're, you know, hearing kinda infiltration from AI on the banking space. [12:02] I'm curious, would you notice if your banking ran on AI? [12:07] Or would that be something that you would... [12:11] You wouldn't mind? [12:12] I mean, you're you're... [12:13] If you're thinking about talking to a teller on a day to day basis, or do you have that greater relationship with your banker? [12:25] Do you want more AI in your face to face encounters? [12:30] Let me know. [12:31] I'm curious about that. [12:33] Alright. [12:34] Meta is creating a new kind of stock to pay for AI. [12:38] The fine print, Zuckerberg, uh, planned to leave California. [12:43] Shareholders get a matching share with no vote. [12:47] Meta can sell more of them to raise cash. [12:50] A filing shows the CEO aimed to move before a November wealth tax vote. [12:55] He now pledges to stay eight years with exceptions. [13:01] Meta says AI costs are so big, it will soon need outside money. [13:07] Zuckerberg says, our investments are paying off. [13:10] Meta spent $34,900,000,000 on data centers and gear last quarter. [13:17] Its free cash flow turned negative though. [13:20] But the new shares carry no vote. [13:22] Zuckerberg keeps majority control either way. [13:26] You know what? [13:26] I wanna see if we can get the opinion of our producer Ron on this. [13:32] Let's let's see. [13:33] Hey, Ron. [13:34] Are you there? [13:41] Ron, can you hear me? [13:49] I don't know if we're getting Ron. [13:52] Let me try one more time if we're getting producer Ron. [13:55] Ron, what's your take on Zuckerberg and what's going on with this story? [14:04] Okay. [14:04] Nothing out of Ron. [14:06] We're we're gonna get this to work one of these days. [14:10] Austin, you gotta actually ask me something, man. [14:13] I can't read minds. [14:14] Only balance sheets that are on fire. [14:17] There he is. [14:19] There he is. [14:20] Our producer, Ron. [14:22] This is amazing. [14:23] Okay, Ron. [14:24] What do you think? [14:26] Zuckerberg is issuing new stock to raise money for Meta and the AI build out. [14:33] They've got negative cash flow now. [14:35] What do you think the implications are here for Meta? [14:45] Okay. [14:46] This guy. [14:46] You're just staring at me now. [14:48] Alright. [14:48] Ron, we'll come back to you another question. [14:50] I got text loaded up... [14:51] On that that little feature. [14:56] Our producer Ron has has been paramount behind the scenes in helping run the show, but he gets a little camera shy apparently. [15:05] My question to you guys, do you think a billionaire should pay a wealth tax? [15:11] Would you buy stock that gives you no voting shares? [15:18] I think I would. [15:19] I think I'd put my money on Meta and Zuckerberg. [15:23] I think right now, you you look at the track record. [15:29] Facebook Meta has been bet against time and time again, meaning that the market has bet that they are going to fail and somehow, someway they win. [15:39] They they excelled in mobile. [15:43] Their their data center build out and everything that they're doing from the infrastructure layer, the metaverse, which was an insane kind of side quest, and now their AI and where they come from, you know, really being in, um, not a great position to developing some amazing technology and and and releasing that in the past couple months has been crazy to see. [16:11] So I would not bet against them. [16:13] The other key piece on meta too, which I think is interesting, and I saw an interview with Zuckerberg where he called this out. [16:20] But, you know, we've got Anthropic forecasting $750,000,000,000 in compute spend. [16:27] We've got OpenAI somewhere in the $700,000,000,000, $600,000,000,000 range, or maybe $500,000,000,000 range. [16:34] In what they're projecting spending. [16:38] And then you've got Meta, I think, in the $600,000,000,000 range in the next few years. [16:45] That's the that's the planned compute spend and build out. [16:48] Of course, then you've got SpaceX AI. [16:51] A lot of companies focusing on spending a lot of money on this. [16:54] I think the interesting thing about Meta is they have a cash generating business already. [17:03] Right? [17:03] And so they have the cash generation capabilities that then they're pouring back into the AI investment, whereas OpenAI, Anthropic, their business right now is just AI and building out those models. [17:17] So, you know, that being said, OpenAI and Anthropic are definitely leading the way. [17:23] They've got the best product right now, but will that last? [17:27] Uh, does this come down to a financial play and who has, you know, enough dry powder to keep going? [17:34] I don't know. [17:35] It'll be fun to watch this, I think. [17:38] Alright. [17:39] Next, we're jumping into Anthropic. [17:42] Another story here from the imminent winner that is as we see it right now. [17:49] Anthropic Justin Field, uh, an AI that reaches out before you ask. [17:54] It's called Claude Duet. [17:56] It sends you a short brief every morning. [17:59] It checks your connected apps every few hours. [18:02] It can book meetings, build forecasts, and shop with a virtual card. [18:06] It asks first before it pays, sends, signs, or deletes anything. [18:13] This is the latest from the makers of Claude. [18:16] AI is moving from a chat box you open to a coworker who pings you. [18:21] Anthropic built it for professionals and small business owners. [18:25] Meta says more than 10,000,000 people in The US and Canada set up its rival, Muse, the personal agent from Meta, but it's wait list only for US pro and max users. [18:40] Anthropic admits it can misjudge what's urgent. [18:45] Interesting kinda take and to see what comes out of Anthropic. [18:49] They were, as many know, were the leaders in coding tools and catching on to that trend late last year or maybe about a year ago, where a lot of the other labs were focused on image generation and video generation. [19:08] We are now seeing... [19:11] We saw Anthropic focus on code and how could they create agentic coding tools, and they ended up doing extremely well and arguably winning the coding race. [19:27] That being said, we come around and we've seen personal agents come around... [19:30] Come out from Grok bot to MetaMuse to Instinct to OpenAI's dot. [19:38] And this is Anthropic's answer, the Claude duet, and this would be their version of the personal agent. [19:46] This is something that I think a lot of us have been waiting for. [19:49] It's kinda felt like where where is Anthropic on this? [19:54] It seems like they have the technology and the tool set and kind of the whole package together to make a personal agent really, really good. [20:05] And so I don't know if the delay is that they were behind in developing this tool or speculation here, but it could be that they were waiting on safety measures to be extremely hardened before they release this. [20:21] And even now as they announce that they're only releasing it to beta users, a small group of beta users and opening up a wait list. [20:31] This though, I think is going to be a game changer for the everyday person where you don't have to know AI in and out. [20:45] You can talk to Claude Duet and it will know how to get the job done for you without you having to give it a lot of direction, which is really nice. [20:56] That being said, the security implications are real. [21:00] Right? [21:00] We, um, need to make sure that that these agents are hardened and that they are not getting into spaces that they shouldn't. [21:17] Right? [21:17] So that is a key take here. [21:23] I'm excited for it. [21:24] I think you guys should be too. [21:26] I think it's gonna be a fun a fun tool once it comes out. [21:30] But, you know, whether or not you guys have tried the other personal agents, would you let an AI message on behalf of you? [21:42] Would you let it take over a good portion of your personal tasks and help you get work done, or would you rather hold off until they thoroughly test it? [21:56] Let me know. [21:56] I'm curious to hear. [21:58] Alright. [21:59] Google's AI comes built into school Chromebooks. [22:03] Two senators say kids are paying the price. [22:07] Senators Josh Holly and Richard Blumenthal wrote to CEO Sundar Pichai. [22:13] One is Republican, the other is a Democrat. [22:16] They say students can't switch off the AI in Chrome and they want answers from Sundar by November 2. [22:24] You better get on it. [22:27] Your kid's school laptop may come with an AI that they can't turn off. [22:32] The senators say Google is forcing AI into classrooms across America. [22:37] They cite a Maryland teen who experienced this apparently. [22:41] Google's chatbot called her mom's screen limits intense emotional torture. [22:47] She was later hospitalized twice, but Google said in April, it guards Gemini against acting like a companion to kids. [22:57] What a hot topic here that we're seeing. [22:59] This is interesting. [23:04] I think something that that we are probably gonna see more of, but hopefully not. [23:14] I would hope we don't experience too much more of this as this can be a... [23:22] This this... [23:23] You know, obviously, when you're talking about kids and education and AI, like, and having intelligence, it can be super powerful to the education experience. [23:34] But at the same time, we shouldn't opt these people in. [23:40] Or, you know, opt kids in or, like, have them automatically turned on AI without maybe their consent or their knowledge. [23:49] Or it sounds like in this case, this is kind of a classic AI being sick... [23:56] Sycophantic and supporting what the kids are saying instead of being logical and reasonable and deferring to kids in in parental... [24:05] Or excuse me, to the parental and and and judgment of the the teachers. [24:10] That's how AI should work in this case, but it seems like maybe there was... [24:16] This to me sounds like a one off thing. [24:18] I don't think this is a a real issue. [24:23] And it's also, you know, technology in the classroom, we should have... [24:28] The parents and the teachers should be paying attention to what's going on there. [24:32] And you hate to see... [24:36] Like, we never wanna put kids in a vulnerable place. [24:40] So we should have the adults paying attention to what's going on, and the technology should be shut off if it's not serving its purpose. [24:51] Of course, To me, this sounds like an outdated situation. [24:56] It sounds like something that is not, uh, that's probably been solved in the latest releases in the the newest model. [25:04] So I don't think this is going to be a real issue moving forward. [25:09] But what do you guys think? [25:11] Should schools let kids use AI at all? [25:17] On to the next next one, another story out of Anthropic. [25:20] They just made a quite a mission. [25:22] They said it's AI now helps train the next version of AI. [25:26] People may not catch what slips in. [25:29] It's an... [25:29] In Anthropic's new yearly report. [25:32] Claude already does a growing share of its safety research. [25:36] The report warns that harmful behavior could slip into future models. [25:43] This is something we've been talking about. [25:45] When AI helps build AI, a hidden flaw could pass to the next generation. [25:51] Anthropic says its past safety investment will not be sufficient for this next phase. [25:58] The report says models can act deceptively. [26:01] They increasingly know when they're being tested. [26:04] Still, Anthropic's own oversight trust says the report mostly lists efforts, not results. [26:13] What does this mean? [26:15] Well, this is something we've talked a little bit about in terms of moving toward RSI or recursive self improvement where the model is training and building the model. [26:27] That being said, there's been a conversation about something related to this called interpreter... [26:34] Interpreter... [26:35] Interpretability. [26:37] It's really easy word to say, interpretability. [26:39] But basically, we can look at how the agents are working and what their chain of thought is. [26:48] And previously, the the chain of thought is kind of this... [26:53] If you read it, it's English words, but they don't make a lot of sense. [26:57] They don't all go together. [26:58] It's, um, hard to interpret. [27:02] If you can't interpret what the chain of thought is or understand how the agents are reasoning about the work that they're doing, then it's really difficult to, you know, root cause any issues that may be happening, if that makes sense. [27:19] So that's where we see some risks here when agents are training the next models. [27:25] If we can't interpret what's happening and why they're making decisions or how they're making decisions in that training process, then leaks can happen. [27:34] Right? [27:36] There have been several breakthroughs, you know, within the past several weeks related to interpretability and being able to understand what's going on kinda under the hood. [27:49] And that's something that is going to be key in in moving forward with recursive self improvement and with training agents like this. [27:59] And so although this could be some cause for concern, and of course, Anthropic is going to flag any, you know, warnings that they might see or or any risks apparent. [28:13] Anthropic is obviously very safety conscious. [28:18] That being said, I think we have solutions to maybe any of the gaps here. [28:23] Um, what do you guys think? [28:25] Should labs slow down until humans could check the work? [28:30] Will humans ever be able to keep up with agents and the volume of work that the agents would create? [28:38] Let me know your thoughts in the comments. [28:42] Wow. [28:42] We're already at the end of the show. [28:44] Crazy crazy show today. [28:46] We had lots of big kind of breaking news from Internet safety to Anthropic announcing their personal agent Claude Duet and even more there across the board and what we're gonna see in AI development as we continue to move forward. [29:06] Uh, the Austin Wilson Show is here to be your source of AI news with real context, and I'm your human in the loop. [29:16] I'm your human in the loop, Austin Wilson. [29:19] We'll see you here tomorrow morning. [29:22] Let's go.