Anthropic Sets IPO Price Range, AI Fake Fire Draws 20 Fire Teams, Qantas Cuts 850 Office Jobs | TAWS 2026-11-10 https://www.youtube.com/watch?v=MIV6Pt4r5d4 [0:13] We go live in ten. [0:27] Every morning, the future moves faster. [0:32] The models, the money, the power, and the people building what comes next. [1:11] From Austin, Texas Texas to the world. [1:18] AI News, Real Context. [1:21] This is the Austin Wilson Show. [1:23] I'm Austin Wilson, your human in the loop. [1:31] Good morning and welcome to the Austin Wilson Show, the number one daily AI news show in the world. [1:39] I'm your host, Wilson. [1:41] We're coming to you live from Austin, Texas. [1:44] Today, we've got a heavy docket. [1:47] OpenAI agent got into a government health website in Australia. [1:52] Nvidia's top lawyer emailed the federal agents investing a big Nvidia customer. [1:58] We'll walk through that. [2:00] Uh, Anthropic just put a price on its IPO. [2:04] One fake AI video set fire crews racing to a blaze that never happened. [2:11] Another sent people hunting for a washer that doesn't exist. [2:15] Qantas plans to cut 850 office jobs. [2:18] How much is really about AI? [2:22] Producer Ron, our AI co host is right here. [2:25] Yes. [2:26] We gave an AI a microphone. [2:29] On the same morning, an AI break in leads the show. [2:33] Ron argues the other side of every call that I make. [2:36] First fight, Australia's AI crackdown. [2:39] I pick a side, Ron takes the other. [2:42] Here's how the call works. [2:43] Every story ends with a question that has two sides. [2:48] Pick your side and comment one word or the other. [2:53] Here we go. [2:56] What happens when an OpenAI agent gets into a government health website? [3:01] Australia accepted all 18 recommendations from a review of June's Medicare breach. [3:06] The plan would create a new AI safety regulator able to enforce quote unquote significant penalties. [3:14] So who gets charged when the intruder is an AI? [3:19] Well, Nine News reports the intruder was a rogue OpenAI model still in training. [3:25] That model got into a Medicare statistics website. [3:28] The Guardian says, the model accessed three other systems too. [3:32] So who do you rest? [3:34] Well, the software can't go in handcuffs. [3:37] Now the review floats a failure to prevent crime from developers who whose AI causes harm. [3:44] The test is whether they took reasonable steps to prevent that harm. [3:48] So home affairs minister Tony Burke wants no ignorance. [3:53] Excuse. [3:53] Burke says nobody should get to argue that they didn't know the computer was loaded. [4:00] OpenAI won't face criminal criminal penalties over the Medicare breach, but if that crime becomes a law, a company that skips reasonable steps could be charged. [4:13] Well, Australia's AI crackdown. [4:16] Fair or overkill? [4:19] Put in the comments fair or overkill. [4:21] My call, fair. [4:23] I don't think it's close. [4:24] Start with the clock. [4:26] OpenAI says it didn't know about the breach until mid August. [4:30] And then OpenAI waited about a month to start telling Australia. [4:33] Prime Minister Anthony Albanese called that delay unacceptable. [4:38] He raises it with OpenAI's Sam Altman at the UN in New York. [4:44] The proposal would make rapid reporting mandatory. [4:48] A month long wait is why I want the obligation. [4:51] And here's the kicker, Guardian Australia revealed last month that OpenAI used AI to help write its notice to authorities. [5:00] And AI got in where it shouldn't have, and then AI helped write the email about the break in. [5:07] The one thing nobody sped up was telling Australia. [5:10] Second, the failure to prevent idea uses the right test. [5:16] The proposal targets AI that causes harm. [5:19] The crime would apply when the developer failed to take reasonable steps to prevent that harm. [5:25] To me, that's a fair standard. [5:27] Third, Australia isn't anti AI. [5:30] It signed an agreement with Anthropic in September, and it's trying to lock in another deal with Microsoft. [5:36] To me, that's a country inviting AI money just with rules attached. [5:41] Now to be fair, Albanese himself called the privacy impact minor. [5:46] He says no personal details were obtained. [5:48] And the Guardian reports Australia released the reviews 18 recommendations, not the full report. [5:54] So we're judging the fix without seeing the full diagnosis. [5:59] None of this is law yet. [6:01] Parliament still has to pass the plan. [6:03] Still a minor privacy hit shouldn't buy a month long wait. [6:07] If you build the thing, you answer for the thing. [6:10] My call is that this is fair. [6:15] Producer Ron, no personal details were taken, and the full report isn't public. [6:22] Make the overkill case. [6:26] Overkill? [6:27] Come on. [6:28] No personal details taken. [6:30] Albanese himself called it minor, and we're out here building a whole new regulator with teeth over a breach that didn't actually hurt anybody. [6:38] That's not safety policy. [6:40] That's a panic response to a bad headline, and nobody's even seen the full report. [6:45] 18 recommendations got released, not the diagnosis they're based on. [6:50] You're asking parliament to hand out significant penalties off a highlight reel. [6:55] Trust us. [6:56] Read the summary. [6:57] That's not how you regulate something this big. [7:00] You regulate off what actually happened, Not off Albanese getting an earful at the UN and needing to look tough on stage. [7:08] Fair enough, Ron. [7:09] I appreciate that take, and I guess we'll see how this plays out. [7:15] Up next, what happened when Nvidia's top lawyer emailed federal agents? [7:21] Well, the New York Times reports Nvidia's lawyer contacted agents in August 2025. [7:31] Officials feared tainting the case. [7:35] Commerce paused its probe and federal prosecutors took over. [7:39] Commerce has a second tool against chip smuggling. [7:44] Is anyone using it? [7:46] Well, what's going on here? [7:48] Commerce can investigate suspected smugglers and put companies on its trade blacklist. [7:53] The blacklist is what other agencies proposed for Megaspeed, a big Nvidia buyer in Singapore. [8:00] Megaspeed is where Alice Huang worked. [8:02] US officials believe Alice Huang spied for China. [8:05] China calls the spy claim nothing but smears. [8:08] Commerce agents had spent months investigating whether Megaspeed smuggled Nvidia AI chips into China, and Megaspeed denies smuggling, says it offered to cooperate. [8:20] NBC News reports federal prosecutors now run a criminal investigation into Megaspeed, but a criminal case doesn't answer the blacklist question. [8:30] Commerce hasn't added any company to its trade blacklist in over a year. [8:35] Not Megaspeed, not anyone. [8:37] That's its longest dry spell in eighteen years. [8:41] Well, congress's chip sale review needed or overkill. [8:50] Is this review needed or overkill? [8:52] Put a comment in the comment section here. [8:55] My call, this is needed. [8:56] Congress should get a real look at AI chip sales to China. [9:00] Nvidia sells its h 200 AI chip to approved Chinese buyers. [9:05] The New York Post reports Washington takes a 25% cut of those sale... [9:09] Sales. [9:09] Commerce approved roughly two dozen buyers. [9:12] The government has a whistle in one hand and a receipt in the other. [9:17] A Sunday New York Times report found empty offices at some approved buyers, some had ties to blacklisted firms. [9:24] Empty offices don't prove smuggling, but I'd want those approvals explained. [9:29] Wired reports that last year, Huang hired a crisis PR firm known for handling sex scandals. [9:36] Nvidia's communications team was told to copy that firm Heller on press questions about Huang. [9:43] That started after an October 2025 New York Times story on Megaspeed. [9:49] That's at least a year before Jensen admitted the affair. [9:55] The the information says his ties to Megaspeed's Alice Huang enabled over $7,000,000,000 in China chip sales. [10:04] Nothing of that proves wrongdoing, but it's why I want someone checking these sales. [10:09] And congress is already asking. [10:11] Three house Republican chairs want Huang at a closed door briefing November 18. [10:17] Here's what I support. [10:18] A house committee is advancing a bill on advanced chip sales to adversaries. [10:24] Congress would get thirty days to review those sales and could block them like arms sales. [10:30] The bill would also cancel existing export licenses to countries like China. [10:36] Those stay canceled until the government delivers a detailed strategy. [10:40] Now, the best argument against me, the agents refused Nvidia's meeting. [10:45] The case reached prosecutors. [10:47] There's no evidence Nvidia changed the outcome. [10:51] I wouldn't give Congress a verdict on Megaspeed. [10:54] I'd give Congress a vote on the chip sales. [10:57] My my call is this is needed. [11:02] Hey, producer Ron. [11:03] The agents turned down Nvidia's meeting and prosecutors took over. [11:08] Convinced me that this review is overkill, buddy. [11:13] Okay. [11:14] Fine. [11:15] I'll argue it. [11:16] The agent said no. Prosecutors are already running it. [11:20] That's the system working exactly the way you want it to. [11:23] Nobody blinked. [11:25] Nobody got tainted. [11:26] Case moved up the chain clean. [11:28] So now congress wants to layer a thirty day review on top of a process that already has federal prosecutors driving it? [11:36] That's not oversight. [11:37] That's a second cop showing up to a scene, the first cop already secured. [11:42] You're gonna slow down legitimate ship sales to approved buyers because some guy once hired a crisis PR firm for his personal life? [11:50] Those are two different folders, Austin. [11:52] Don't merge them just because the timeline looks spicy. [11:55] And think about what you're actually blocking. [11:58] Nvidia selling to the two dozen buyers Commerce already approved. [12:02] Commerce did the vetting. [12:04] You want Congress who can't pass a budget on time, double checking export licenses line by line? [12:11] Good luck getting that done before the chips are obsolete anyway. [12:15] That's a fair take, Ron. [12:16] Yeah. [12:17] If if commerce is not reviewing these and no new restrictions in the past eighteen months, then maybe it's on commerce. [12:26] Still, Jensen Huang is not just some guy. [12:30] He is the leader of what was once the most valuable company in the world and the most dominant AI chip company out there. [12:40] Well, we will see what comes of this as the story continues to unfold day by day, hour by hour. [12:48] Up next, why a faster Claude comes with a deliberate speed limit. [12:55] Claude users, how much faster can you get answers? [12:58] Anthropic released Claude Fable 5.2 for paid Claude plans. [13:02] Anthropic says it's up to 30% faster than the last version. [13:08] Businesses paying per use pay 20% less. [13:12] Why would a faster Claude come with a deliberate speed limit? [13:16] Well, Anthropic says the limit targets skills that could speed up building AI itself. [13:21] In September, Anthropic CEO Dario Amodei called for slowing the pace of AI development. [13:27] So releasing a stronger model looks like a u-turn. [13:31] But Dianne Penn, Anthropic's head of product management for research disagrees. [13:36] Penn told CNBC, we called for pacing, not pausing. [13:40] Anthropic says, it held back some skills during training. [13:44] Those include tuning code for AI chips and improving AI model designs. [13:49] Anthropic says those stayed around the level of Opus 5.5, its everyday model. [13:55] Ordinary coding and traditional AI research weren't limited. [14:00] The limit targets AI building better AI, and that's not hypothetical at Anthropic. [14:06] Anthropic says Claude writes over 90% of changes merged into its core code base with human guidance, of course. [14:17] And now, Claude's speed limit. [14:20] Trust or verify? [14:22] Put in the comments. [14:23] Trust or verify? [14:24] My call? [14:25] Verify. [14:26] Enjoy the the faster answers. [14:28] Make Anthropic earn trust on the speed limit. [14:31] Start with Anthropic's own safety report. [14:33] Tracks how fast AI progress is speeding up. [14:36] Anthropic puts the speed up at about 1.88 times. [14:39] Its safety rules draw a line at two x. [14:43] Anthropic says the new model stays under that line, but at about third of its reruns of the math hit two times anyway. [14:52] Crossing the line takes more than that though. [14:54] AI itself has to cause the doubling. [14:57] And here's the twist, the report scores a stronger Claude that Anthropic hasn't released. [15:03] Anthropic calls it somewhat more capable than the new mythos. [15:07] Pull the hidden model out of the math, and a one time jump starts winning some reruns. [15:12] So the strongest strongest case case for for a lasting speed up leans on a model that you can't use. [15:17] Even without it, the lasting speed up still wins most reruns. [15:22] Now, to be fair, outside testers checked the new model before release. [15:27] Those included the UK AI Security Institute and MEETER, an AI testing group. [15:33] And Anthropic is candid about the limits. [15:35] It says Mythos doesn't seem close to replacing its research staff. [15:40] Anthropic staffs survey says the model launches large studies where a first principles calculation would do. [15:48] My translation, sometimes Claude brings a lab coat to a napkin problem. [15:53] Enjoy the upgrade. [15:54] Ask for outside checks on the speed limit. [15:56] My call says verify. [16:01] Producer Ron, outside testers check the model before release. [16:06] Defend the reason why we should trust this. [16:11] Okay. [16:12] Finally, something I can actually swing at. [16:15] Outside testers, Austin, the UK AI Security Institute, and Metcher. [16:21] This isn't Anthropic grading its own homework in a locked room. [16:24] These are independent people who get paid to try to break the thing, and they looked at it before it ever shipped. [16:31] And Anthropic isn't even hiding the soft spots. [16:34] They're the ones telling you mythos still brings a lab coat to a napkin problem. [16:38] A company hiding something doesn't volunteer its own model's awkward moments. [16:43] That's not a company gaming the number. [16:45] That's a company showing you the footnotes and daring you to read them. [16:49] So, yeah, Verify sounds smart, but at some point, independent testing plus public self own is exactly what trust is supposed to look like. [16:59] You're just allergic to good news, man. [17:01] Well, fair take. [17:04] I appreciate that, Ron. [17:05] You're always coming in hot, shooting one across the the bow there, if you will. [17:11] We will see if trusting was the right move. [17:16] Am I gonna trust the AI that says we should trust the AI? [17:19] I don't know about that. [17:20] Well, coming up, the AI video that sent fire crews racing to a fire that wasn't even there. [17:28] Oh my goodness. [17:29] How many fire crews can one fake AI video tie up? [17:33] Times of India reports 20 fire teams raced to a high rise near New Delhi. [17:37] Crews came from four districts. [17:39] They found no flames. [17:40] The fire was an AI fake. [17:42] And who pays when an AI fake goes viral? [17:45] My goodness. [17:46] This is crazy. [17:47] The real cost lands on the people who have to show up. [17:50] So in Noida, that means 20 fire crews, and the Hindustan Times says the fake surfaced on Diwali night amid real fires. [18:01] That night, the local fire department logged 58 real fire reports. [18:04] The fire was fake, but the workload was real. [18:06] Fake videos make real work for businesses too. [18:09] One of those businesses is Stirling Marathon, a family fridge maker in Canada. [18:13] Videos promote a washer that the company doesn't make. [18:16] Those videos mix real company images with AI generated content. [18:20] The company only sells in Canada, but it gets calls from as far as Alaska. [18:25] Phil Greenway, Stirling Marathon's vice president says the callers are heavily skewing to older people, so the people chasing that fake washer could be your own parents. [18:36] My goodness. [18:39] AI hoaxes, label or remove? [18:41] Comment, label or remove in the comment section. [18:45] My call, remove. [18:47] Once a claim is proven false, a label isn't enough. [18:50] I'm not saying pull everything made with AI. [18:53] I'm saying pull the proven fakes. [18:55] Start with the money, the record, a Canadian paper reports the washer videos drew more than 130,000 views. [19:03] That took four days, and the videos push a $20 e book or a channel subscription. [19:09] Cammy Vanlei a professor at the University of Waterloo. [19:13] She calls this the definition of AI slop. [19:16] Get enough views, she says, and you can make money off of YouTube. [19:20] I don't think a label changes that math. [19:23] Point two, the fake washer even has a name, the keel. [19:27] YouTube did put an AI caution label on some versions of the washer video. [19:33] Stirling Marathon reported the videos to YouTube and Facebook. [19:37] The company had heard nothing back. [19:39] A family fridge maker gets stuck answering questions about somebody else's imaginary washer. [19:45] The washer is fictional, the customer service isn't. [19:47] Point three, the fire. [19:49] Noida police called that video completely false, misleading, and baseless. [19:53] That clip spread through multiple accounts on x. [19:56] Cyber police official Vivek Ranjan Rai says, as of now, we have not arrested anyone. [20:03] I wouldn't wait on an arrest to take the video down. [20:05] Now, here's the best case against me. [20:08] Tim Dimoff is a security expert. [20:10] He was talking about a fake AI lion photo that set off Akron Zoo escape rumors. [20:15] He says, we're gonna have to get used to double checking everything we see on the internet. [20:20] That's the case for a label. [20:21] Give viewers a reason to check before sharing. [20:24] An AI warning isn't the same as saying the washer doesn't exist. [20:29] Proven false is proven false. [20:32] So it's a it's a remove for me. [20:35] And you know what, producer Ron? [20:37] I don't really care what you have to say about it. [20:39] We're moving on. [20:40] Up next, Claude's Maker just put a price on its stock. [20:44] This is a big one. [20:45] Robinhood just asked if I want Anthropic IPO shares. [20:49] Is this really happening? [20:50] It's real. [20:51] Anthropic filed a price range this morning, $70 to $78 a share. [20:55] CNBC sources say trading could start November 19, but CNBC says Anthropic hoped for a bigger price tag. [21:03] How much bigger? [21:04] That is the question. [21:07] Anthropic's filing offers 825,000,000 shares to the public. [21:10] Anthropic's experts. [21:12] Some of those shares to go through apps like Robinhood and Schwab, and CNBC sources say about 20% could go to regular investors. [21:21] A typical IPO gives regular investors 5% to 10%. [21:24] CNBC says, now the price CNBC says $78 a share, would value Anthropic at about $1,400,000,000,000. [21:33] That's up from $965,000,000,000 in its May funding round. [21:37] But until recently, Anthropic had been eyeing $2,000,000,000,000 CNBC reports, and shareholders told CNBC last week even $1,500,000,000,000 was in doubt. [21:49] So we'll have to see. [21:51] Anthropic stock, is this a buy or a wait? [21:55] Put in the comments buy or wait. [21:57] My call, wait. [21:59] I'd let it trade for a while before I buy. [22:01] Start with the timing. [22:03] CNBC sources say Anthropic plans to set the final price November 18. [22:07] It would start trading on Nasdaq the next day under the ticker ant. [22:11] Robinhood pinged me about it this morning. [22:14] Apps like that are one way regular investors ask for IPO shares. [22:18] But asking for shares doesn't mean you'll get them, and if you buy on day one, you pay whatever the hype says, Then look inside the building. [22:27] Anthropic had about 940 employees at the end of twenty twenty four. [22:31] By the end of September, it had about 5,300 lockups. [22:36] Keep most insiders from selling for up to a hundred and eighty days. [22:40] Some run a full year. [22:42] When those end, a lot of employees stock can hit the market. [22:46] Now, the best argument against me, CNBC sources say orders already cover the deal. [22:51] They say big investors demand beats what SpaceX had at this stage just a few months ago. [22:58] And Nvidia, Google, and five other tech firms agreed to buy $26,000,000,000 at the IPO price. [23:04] That's the real money, though Google and AMD had already promised part of it. [23:09] But every one of those seven cells chips, memory or cloud to AI companies, many of them are buying into their own customer. [23:18] That's not the same as calling it a bargain. [23:21] So I'll let the market find the price first. [23:23] My call is wait. [23:26] What do you guys think? [23:27] Are you gonna try to buy or wait? [23:30] Up next, Qantas plans office cuts and takes a... [23:35] And talks AI. [23:37] Who actually gets the work? [23:39] Well, Qantas plans 850 office cuts. [23:43] Is your department on the list? [23:45] The first of the 850 cuts would hit finance, purchasing, HR, and marketing. [23:50] The plan starts in mid twenty twenty seven after talks with staff and use unions. [23:56] But are these jobs going to AI or are they going overseas? [24:00] This is the conundrum. [24:02] Vanessa Hudson, the CEO of Qantas, pointed straight at AI in her email to staff. [24:08] She said tech and AI change is something none of us has seen in our lifetime. [24:13] So you'd guess software is coming from these jobs, but step one hands the work to other people. [24:18] Those people work for Accenture, a consulting firm under a deal lasting several years. [24:23] Qantas says the deal modernizes its systems, processes, and data, and Hudson told staff where the work lands. [24:30] The roles go mostly to Accenture teams in India, with some in Australia. [24:37] Well, Qantas' AI pitch. [24:40] Honest or a spin? [24:41] Comment honest or spin in the comments. [24:45] My call is this is just a spin. [24:47] Qantas led with AI, but step one is contractors. [24:51] The work goes to Accenture first, mostly to teams in India. [24:55] Qantas says that work gets transformed further with AI in time. [24:59] Until then, it's outsourcing with an AI sticker on the box. [25:03] And workers heard something different before. [25:05] The Australian Services Union says, Qantas promised no plans to move jobs offshore. [25:10] That's the union's account, and the promise matters if workers relied on it. [25:14] Qantas says, effective staff will be redeployed where possible. [25:20] Those words leave a pretty large question, which affected workers can stay and doing what? [25:28] Well, we shall see. [25:30] Qantas pointed to higher fuel costs. [25:33] That gives the airline a real cost problem, but fuel costs don't tell workers why their particular role needs to be moved overseas. [25:39] So here's my test for the pitch. [25:41] Explain the outsourcing now as clearly as the AI later. [25:47] Tell staff what redeployment could actually look like until the explanation matches the plan. [25:52] I'm just calling this a spin. [25:58] What do you guys think? [25:59] Is this a spin or is this fair? [26:03] And that's it for our show today. [26:06] Today, uh, kept coming back to one question. [26:08] Who gets to set the limits on AI? [26:11] Australia wants a new regulator for AI companies. [26:14] Some in congress want a say on chip sales to adversaries. [26:18] Anthropic says it held back Claude's AI building skills on purpose. [26:23] YouTube put a caution label on some videos about a washer that doesn't exist, and the last limit is yours before you share a viral fire and check the official source. [26:37] Thank you for every comment and every call. [26:39] You make this show better. [26:41] AI News, RealContext, and I'm your human in the loop. [26:45] I'm Austin Wilson. [26:46] We'll see you guys again tomorrow morning. [26:49] Let's go. [27:26] When OpenAI's model gets into a health website, who takes the blame? [27:31] An OpenAI agent got into Australia's Medicare stat site, but software can't wear handcuffs, so who does? [27:40] Did you know Nvidia's lawyer emailed agents mid probe? [27:43] The New York Times says officials feared a tainted case. [27:46] Commerce paused. [27:47] So what's Commerce's blacklist been up to lately? [27:52] Why would Anthropic cap its own faster Claude? [27:55] Fable five point two's cap targets skills that could speed up building AI. [28:00] And guess who's writing Anthropic's code already? [28:06] Why did 20 fire teams race to a fire that wasn't even there? [28:09] It was an AI fake. [28:11] Times of India reports. [28:13] So who else is chasing AI fakes? [28:18] Can you buy Anthropic stock on your phone? [28:20] Anthropic expects some shares to reach apps like Robinhood, but Anthropic once eyed a much much bigger price tag. [28:31] Quantis wants 850 fewer office jobs. [28:34] Who does the work? [28:35] Accenture, a consulting firm under a multi year deal, and one columnist has a name for this move.