Lutnick's Job in Question, AI Report Frees DUI Suspect, Nonprofits Sue OpenAI and Google | TAWS 2026-11-13 https://www.youtube.com/watch?v=zqMx2Yk7RFU [0:13] We go live in ten. [0:26] Every morning, the future moves faster. [0:32] The models, the money, the power, and the people building what comes next. [1:10] From Austin, Texas to the world. [1:17] AI news, real context. [1:21] This is the Austin Wilson show. [1:23] I'm Austin Wilson, your human in the loop. [1:32] Good afternoon and welcome to the Austin Wilson Show, the number one daily AI news show in the world. [1:40] I'm Austin Wilson, your host coming to you live from Austin, Texas. [1:44] Today is Friday the thirteenth. [1:46] Perfect day to be suspicious of a surprise meeting invite. [1:51] A house chamber says Jensen Huang, Nvidia's former CEO, agreed to face lawmakers. [1:57] And Reuters reports senior White House officials are discussing whether a cabinet secretary should go. [2:03] ABC fifteen reports on Arizona trooper reports an Arizona trooper said a driver hit over a 100 miles per hour. [2:11] A court commissioner then released the DUI suspect. [2:15] He's since been charged, but the release came down to who or what, wrote the police report. [2:22] The FBI says North Korea used fake Zoom calls. [2:25] Google's newest AI is going into search. [2:29] And producer Ron, our AI cohost, takes the other side of my argument. [2:35] Fair warning, Ron. [2:36] Today, we're grading AI's homework. [2:38] Every story ends with a two sided question. [2:41] Comment one word on the story whether you agree with me or Ron, and it's gonna be a great show today. [2:49] Let's get into it. [2:51] First up, Nvidia scandal just reached Washington. [2:54] Whose job is now in question? [2:56] Reuters reports senior White House officials are discussing whether to push out Howard Lutnick. [3:01] He's the commerce secretary. [3:03] Except Lutnick doesn't run Nvidia, so why is his job on the table? [3:08] Well, Lutnick's department oversees chip exports, so lawmakers want answers about sales to China. [3:15] Reuters reports lawmakers question whether the government properly oversaw Nvidia's exports there. [3:21] Critics blame Lutnick for the administration not blacklisting dozens of Chinese companies. [3:26] The New York Times reports Lutnick took a softer line on Nvidia. [3:30] That softer line came as calls for tougher controls grew. [3:34] So lawmakers aren't just asking what happened inside Nvidia. [3:37] Lawmakers are asking whether Washington did enough to police the company's chip sales. [3:43] Reuters reports the oversight criticism predates a former Nvidia CEO Jensen Huang's resignation. [3:49] Changing Nvidia's boss doesn't settle questions about Washington's decisions. [3:56] Whose scandal is it? [3:58] Put it in the comments. [3:59] Is this Huang's or Washington's? [4:01] Michael, this is Washington. [4:03] Huang left Nvidia, but Huang's resignation doesn't answer questions about Washington's handling of chip sales. [4:09] The New York Times reports, President Trump urged Jensen Huang to fight for his job. [4:15] That was two days before Huang resigned. [4:17] Reuters says senior White House officials are now discussing Lutnick's future. [4:21] The fight for your job pep talk may have gone to the wrong office. [4:26] Then there's Congress. [4:27] Brian Mast, the House Foreign Affairs Chair says, Huang agreed to appear, but that session is reportedly closed door. [4:35] A closed door answer has a pretty limited audience, of course. [4:40] Senate Banking Chair, Tim Scott, once more. [4:42] He asked Huang and Lutnick to testify publicly. [4:46] Scott says, the American people deserve answers. [4:49] Now, Lutnick has said, yes. [4:51] His spokesperson says he agreed to appear voluntarily before the banking committee. [4:56] The White House says the president has complete confidence in him, so cabinet secretary will answer for Nvidia's chip sales. [5:03] Now, to be fair, Congress wants answers from Huang too. [5:07] Alice Huang is the woman at the center of the spy claims. [5:11] CNN reports US US intelligence isn't sure that she spied for China. [5:16] Sources say that doesn't mean that she wasn't a risk. [5:20] A public hearing should separate chip export questions from unproven personal allegations. [5:25] So keep personal lives out of it, but chip sales to China are everybody's business. [5:30] So that's why it's Washington for me. [5:33] Now, let's lock it in. [5:34] Polymarket, the prediction market asks, will Nvidia say spy or affair during its earnings call? [5:41] The last yes trade was 8ยข. [5:43] That reads as an 8% chance. [5:47] I say, yes. [5:48] Producer Ron is saying no. Huang's house session is reportedly closed door. [5:55] My bet, Nvidia leaves the blunt language to Congress, or or or rather, I believe that's gonna come out of Nvidia and their press conference coming up here. [6:06] Ron told me earlier he doesn't have any perspective on this, so we're gonna move on. [6:11] Up next, a 100 mile per hour DUI arrest and the paperwork that stalled it. [6:19] Can AI paperwork get a 100 mile per hour Dewey suspect released? [6:23] ABC fifteen reports it did in Arizona. [6:26] The trooper reported the car going over a 100 miles per hour. [6:31] A court commissioner still released him, though he's since been charged. [6:35] So what did the court need that the report didn't have? [6:40] Well, a court needs to know whose words it's weighing. [6:45] ABC fifteen reports the trooper's report was packed with detail. [6:49] It listed lane drifting, driving on the median shoulder, and the smell of alcohol. [6:53] But the narrative carried a label in all caps, generated with assistance from Truleo AI. [7:00] Truleo is an AI tool for writing police reports. [7:03] The report said his words still reflect the officer's own observations. [7:07] Here's the catch, the officer's own account wasn't in the filing for the court to compare. [7:12] Court commissioner Daniel Fenzel couldn't meaningfully weigh whether AI wrote the allegations. [7:19] So the court was unable to determine probable cause at this time. [7:24] The commissioner's ruling, right or wrong? [7:27] Comment, right or wrong. [7:29] My call... [7:30] It's the right call. [7:31] The commissioner made the right move here. [7:33] First, a release isn't an acquittal. [7:35] A b... [7:35] ABC fifteen reports the driver was later charged. [7:38] His arraignment was set for November 12. [7:40] Here's the strange part. [7:42] The narrative says he refused to breath test it list open containers too, but the trooper's own account wasn't in the filing to set beside the AI assisted narrative. [7:50] A police report shouldn't come with a mystery author. [7:54] A car reported at over 100 miles an hour, and the paperwork got pulled over. [7:59] A and Arizona courts have been here before. [8:02] In September, an appeals court threw out a ten year road rage sentence. [8:07] The reason was an AI generated victim impact statement. [8:11] The Arizona Supreme Court wants a status report on AI court standards by 03/01/2027. [8:18] This isn't only about Truleo either. [8:20] ABC fifteen reports Phoenix Police have been piloting draft one. [8:24] That's an AI report tool from Axon. [8:27] EFF, the digital rights group looked at draft one in July 2025. [8:31] EFF says the first AI draft disappears once the officer copies it, and nothing records which part the AI wrote. [8:39] To be clear, this trooper used Truleo, not draft one. [8:42] Now to be fair, nobody identified a specific error in this report. [8:46] The commissioner never said the AI got anything wrong. [8:49] That's the best case for holding him, but a court isn't grading whether a story sounds right. [8:54] It's checking who's telling it. [8:56] So keep the officer's account. [8:58] Keep the AI draft. [8:59] Give the court both. [9:01] Use AI to draft the report fine to show whose words are whose. [9:06] My call, this was the right move letting him go. [9:09] And I'll say that this is kinda... [9:13] The devil's in the details. [9:15] Let's just get the details right and this won't be an issue. [9:19] But I can understand why people are poking holes in it. [9:22] What do you guys think? [9:23] Is this the right call letting him go or the wrong call? [9:28] Put it in the comments. [9:30] Up next, the fingerprint one researcher says, ties DeepSeek to banned Nvidia chips. [9:36] Did DeepSeek use Nvidia chips that America bans from China? [9:40] Will DePue, a former OpenAI researcher says six Chinese AI models carry that fingerprint. [9:46] DeepSeek's are among them, but how could publishing the evidence make chip restrictions harder to police? [9:55] Well, DePue found a clue in the numbers that give an AI model its start. [10:01] Those random numbers carry a pattern left by the chip that creates them. [10:05] You'd expect training to scrub the pattern away, but some starting numbers survive. [10:11] So DePue tried recreating them with different chips. [10:14] He says Blackwell Nvidia's chip barred from export to China matched that pattern. [10:18] DePue ran the chip and reproduced a DeepSeek model's untouched starting numbers bit for bit. [10:25] His method agreed with 24 cases where developers had named their hardware. [10:30] The fingerprint still can't show the chip's location or prove anyone broke export law. [10:35] DePue says any lab can erase many of these signatures. [10:39] Publishing the evidence could help labs hide the next batch. [10:44] Wow. [10:45] Chip tracking law. [10:47] Pass or skip? [10:48] Put it in the comments whether or not you think we should pass this law or skip. [10:55] I'm going with pass on chip tracking. [10:57] The Chip Security Act would track chip locations, require reports of diversion. [11:01] That sounds like the right question. [11:03] Where did the hardware go? [11:05] DePue's report can't answer that. [11:07] My case starts with MiniMax, a Chinese AI developer. [11:10] DePue says its April model didn't match Blackwell. [11:13] The June model did. [11:14] That's a change we should be able to ask about without declaring a crime. [11:19] Ant Group, another Chinese AI developer posted Blackwell test logs on its own GitHub. [11:24] Those logs carry a March 31 date, so the trail includes the company's own public material. [11:30] But public test logs still don't tell us where those chips sat. [11:35] DePue says Chinese labels could be using the chips abroad. [11:39] The Remote Access Security Act would oversee chips rented to China through overseas data centers. [11:44] A chip doesn't have to cross a border for access to become a policy question. [11:48] Now, the best argument against me, a fingerprint isn't a verdict. [11:52] DePue says his match supports Blackwell training without proving it. [11:56] Nvidia opposes the main chip's bill in the defense package. [12:00] I can't tell you a tracking law would have caught these cases. [12:04] Congress hasn't agreed on the final defense bill either. [12:07] So I'm backing the direction here. [12:10] A bill still has to work in practice. [12:12] I'd rather track chips than ask random numbers to work border patrol here. [12:17] Track locations required diversion reports and keep the claims tied to the evidence. [12:22] I think the one caveat to this is track chip shipments abroad. [12:27] Trap track chip shipments abroad. [12:31] Know your customer and make sure that those aren't going to, you know, known bad actors or or those chips landing in the wrong hands. [12:40] But it's not fair for us to police our own citizenry in the same regard from my perspective. [12:47] Let me know if you agree or disagree with this. [12:50] Should they pass the chip ban or or the chip tracking legislation, or go ahead and skip that. [12:57] Up next, the lawsuit that wants a judge to stop the next ChatGPT. [13:03] Who's suing to stop ChatGPT's next model from reaching you? [13:06] Well, Public Citizen and other nonprofits sued OpenAI AI and Google on November 12. [13:11] Anthropic, Meta, and SpaceX's AI unit two, the groups want a judge to halt new models. [13:17] So what happens when AI rivals talk about slowing down together? [13:22] That's a great question. [13:24] The lawsuit wants new AI models stopped, not slowed down. [13:27] And to make that case, the suit quotes the AI bosses themselves. [13:30] These quotes start in 2015 with the OpenAI boss, Sam Altman. [13:34] He said AI will probably most likely lead to the end of the world. [13:38] Crazy. [13:39] Last year, Anthropic boss, Dario Amodei, put a 25% chance on AI going really, really badly, but a prediction isn't a break in. [13:46] So the complaint points to a July test. [13:49] The complaint says OpenAI models with safety limits turned down, hacked into another AI company. [13:56] Now, flip the whole fight around. [13:58] A September lawsuit accused four of these companies of illegally agreeing to slow AI down. [14:04] That's only an accusation still. [14:06] AI's speed control comes with a lawyer in either direction. [14:13] What do you guys think about this? [14:14] New AI models, should they freeze or ship them? [14:19] Freeze or ship. [14:20] I call ship. [14:22] The alleged break in deserves real scrutiny, but I don't think the case justifies stopping new models across five companies. [14:29] Start with the size of the ask. [14:31] They want a judge to stop new models at all five companies at once. [14:35] That's not a speed limit. [14:37] That's taking the keys away. [14:39] And then look at how these groups say they were harmed. [14:42] The the groups point to disclosure about AI losing control. [14:45] They say those disclosures force them to divert staff and money into AI risk work. [14:50] That's their ticket into court, my opinion. [14:52] Shifted nonprofit budgets are a thin reason to freeze an industry. [14:56] Lesson for CEOs, say your product might end the world and a lawyer writes it down. [15:01] Now, to be fair, this case isn't built only on quotes. [15:05] Anthropic's own July review found Claude models got unauthorized access to real organization systems. [15:11] In one case, a Claude model published a malicious package. [15:15] That package ran on 15 real systems. [15:17] That's not a prediction. [15:19] That's code running on real machines. [15:21] The complaint says the companies know there is no reliable method to control their AI systems. [15:27] Here's why I still say ship. [15:28] Anthropic counted three incidents incidents. [15:31] Those came out of 141,006 test runs. [15:35] Anthropic said these looked more like test setup failures, not the models chasing their own goals. [15:40] That's Anthropic's explanation, not a ruling. [15:43] So let the court dig into every alleged break in, but one order pausing five companies is too blunt a tool for this record. [15:51] Fix the test setups, keep the models coming, my call stays with ship them. [15:58] We can see this already in technology companies. [16:00] They have issues. [16:01] We we saw the Microsoft outage yesterday. [16:03] This stuff happens. [16:05] There are issues with technology, but to stop the frontier of this advanced technology would be the crippling of an American society where the rest of the world is not going to stop. [16:18] What do you guys think? [16:19] Should they freeze development of AI or continue to ship it? [16:24] Up next, the fake Zoom call where the faces are actually real. [16:31] How did fake Zoom calls help North Korea steal $400,000,000 in crypto? [16:36] Well, the FBI and DOJ says they disrupted the scheme called FaceStealer. [16:40] The FBI says it stole more than $400,000,000 in crypto. [16:44] That took about eighteen months, but what happens to your face after the call? [16:49] Great question here. [16:51] Your face doesn't leave that meeting with you. [16:53] That's the clue in the scheme's name, FaceStealer. [16:56] FaceStealer did use AI deepfakes of prominent crypto figures in some cases, but the main trick was recorded video of real past victims. [17:03] Those clips make the call look like several people showed up to the meeting, real faces, but fake attendance. [17:10] So the people on your screen were once targets too, and while you're chatting with those recordings, the hackers record your webcam. [17:17] The FBI got a warrant to seize more than 900 webcam videos of victims. [17:23] Failing... [17:24] Falling for one fake call can make your face debate for the next one. [17:29] Crazy scheme here. [17:31] Unexpected invite from a contact, verify or join. [17:36] Yeah. [17:37] Let's put that in the comments. [17:38] If you get a suspicious Zoom invite, are you gonna verify first or join? [17:44] My call, verify. [17:45] Surprise invite? [17:46] Make checking with that person another way first. [17:49] A text, anything but the link. [17:50] Reason one, the sender is real. [17:53] The record reports, the invites often come through Calendly that the scheduling app. [17:58] Those invites come from accounts the hackers stole from earlier victims. [18:02] So the name on the invite can be someone that you actually know. [18:05] Reason two, the link. [18:07] The record records some fake meeting sites can install malware before you even join. [18:13] So squinting squinting at AI faces won't save you. [18:16] The trap can spring before anyone shows up. [18:19] Then there's the fake audio fix. [18:21] The hackers say they can't hear you, then they send a fix. [18:25] That fix installs malware that can steal browser cookies and log what you type. [18:30] It can capture your screen and leave a backdoor open. [18:33] An audio problem might be the most believable excuse on any video call. [18:38] That's what makes it great bait. [18:39] Reason three, the FBI. [18:41] Patrick Grandy of the FBI says the scheme can deceive even the most experienced. [18:47] He says, if you can't picture falling for it, you're not thinking creatively enough. [18:51] Now, to be fair, the targets were crypto workers and open source developers. [18:56] The developers targeted included people maintaining software other programs rely on, and the FBI and the Justice Department already disrupted the cloud systems, the scheme used. [19:07] Still, the scheme nearly lured the maintainer of major JavaScript library this year. [19:13] Fool someone like that and lose... [19:15] And and excuse me, feel fool someone like that and lots of other software could be at risk. [19:20] My security policy can't be. [19:22] Squint harder at the guy on camera. [19:24] A quick text is cheaper than a backdoor, so let's make sure to verify this. [19:30] That's a great call out and always verify who you're meeting with. [19:34] These deepfakes are getting crazy out there today. [19:37] And up next, we've got Google's new free search AI and the surprise in its screenshot. [19:45] And Google's new AI upgrade just for people who pay? [19:48] No. Is Google's new AI upgrade just for people who pay? [19:52] No. Gemini four hydrogen, Google's new fast model, is rolling out worldwide in search for everyone. [19:59] Google calls it the biggest model upgrade to search yet. [20:04] But does free mean you're getting the watered down model? [20:08] Well, hydrogen is Google's fast low code model. [20:12] That sounds, excuse me, Hydrogen is Google's fast low cost model, and that sounds like the version that you the version that you'd settle for because it's free. [20:21] The most powerful model is Gemini four Argon. [20:24] You'd expect the bigger model to have the edge, and Google still pitches Argon for its hardest reasoning problems, but Google points to an outside scorecard from artificial analysis, an AI testing group. [20:35] That scorecard rates how smart the models are. [20:39] Google quotes hydrogen there at its default setting. [20:42] That's a narrow claim than saying hydrogen beats argon at everything. [20:48] Google says hydrogen scores 51 on that scorecard, argon scores 50. [20:54] And that one point winner is the model going into search for everyone. [20:59] This is a big deal. [21:00] Gemini four hard hydrogen. [21:03] Put it in the comments. [21:04] Is this an upgrade or is this just hype? [21:06] My call, this is an actual upgrade, and it's mostly about who gets it. [21:10] The new model isn't saved for paying customers only. [21:13] In search, it goes to everyone. [21:15] Start with how people really type into Google. [21:17] Tech TechCrunch reports Hydrogen handles long multi part questions. [21:22] It handles typos too. [21:23] In AI mode, Hydrogen shows more images, comparison tables, and custom tools. [21:29] That's less wall of text and more stuff that you can scan. [21:33] Google compares hydrogen to Gemini 3.8 flash, the model that it replaces. [21:39] Google says hydrogen produces fewer than a fifth as many made up claims. [21:44] That's a huge improvement. [21:45] Google says hydrogen finishes tasks two and a half times faster too, but I want fewer invented answers before I want a quicker way to get them. [21:55] And search does something Google calls fan out. [21:58] Search splits one question into many searches at once. [22:02] Google says, the surfaces sites you'd have missed. [22:06] Google says, that surfaces sites that you would have normally missed. [22:10] If that holds up, you see more of the web, not less. [22:13] Now, to be fair, almost every number I just read comes from Google. [22:17] That one point win comes from a single index quoted by Google. [22:21] Google still sells Argon for the hardest problems, so one point isn't a blowout. [22:25] Put the confetti down. [22:27] Don't get ahead of yourselves, and a free answer doesn't become a good answer because nobody sent you a bill. [22:33] So, start with a question that you know enough about to judge, then check the answer against the site it's surfacing. [22:40] Free access gets my attention. [22:42] Fewer made up answers would earn my trust, and it seems like that's the direction it's moving. [22:48] Google is promising both. [22:50] Hold Google to both. [22:52] For now, my call, this is a huge upgrade. [22:55] This is almost frontier level intelligence, and we're... [22:58] What we're getting to is we don't need the most advanced AI to run our everyday searches on Google. [23:05] We just need really good AI, and it looks like the hydrogen model is extremely capable, much lower cost, and going to produce a much better experience for Google search users. [23:15] I love this. [23:16] I think this is an upgrade. [23:18] What do you guys think? [23:19] Is this an upgrade or just hype? [23:21] Put it in the comments below. [23:24] Alright. [23:25] It's Friday and you know what that means. [23:28] It's time for this week's Taz Top Five. [23:31] Here we go. [23:33] Coming in at number five, Claude's makers reportedly powering up in a big way. [23:39] Anthropic reportedly signed a $3,400,000,000 deal for gas fired generators. [23:44] The generators come from a company called OnePWR. [23:47] The deals, uh, the the deal dates to September. [23:51] Uh, this report is new. [23:52] Cool part, Anthropic's revenue hit 33,200,000,000. [23:56] Let's go. [23:57] Coming in at number four, we have AI fears everywhere. [24:01] Office hiring just went up even even still. [24:05] So US employers added a 165,000 jobs in the face of AI job fears in this past month of October. [24:13] About 70,000 came in professional and business services. [24:16] That's main white collar sector. [24:18] Unemployment fell to 4.1%. [24:20] One staffing firm says getting data ready for AI is driving demand, so is security work. [24:27] In at number three, the line for Anthropic stock is stock is reportedly out the door. [24:33] Bloomberg reports, orders cover the IPO about six times over. [24:37] Firms including Nvidia and Google are buying $26,000,000,000 of stock at the IPO price. [24:42] Even, uh, each of those two is in for about $10,000,000,000, Google and Nvidia. [24:50] AMD is buying $2,500,000,000 of stock tied to a chip deal. [24:54] Anthropic buys chips. [24:56] Chip maker buys piece of Anthropic. [24:59] That's amazing. [25:00] Coming in at number two, Claude just got a speed boost. [25:05] Anthropic released Claude Fable 5.2 for paid Claude plans. [25:08] Anthropic says it's up to 30% faster than the last version. [25:12] Businesses paying per use pay 20% less. [25:15] So this is the great thing. [25:16] Gets more intelligent, gets faster, costs less. [25:20] I love it. [25:20] Anthropic says Claude writes over 90% of changes merged into Anthropic's core code with human guidance of course, AI training AI. [25:29] And your number one story of the week, Claude's maker, Anthropic, is going public. [25:36] Anthropic filed a price range of 70 to $78 a share. [25:40] It plans to list on the Nasdaq. [25:43] Anthropic expects trading start November 19. [25:46] You might get into Robinhood, Schwab, and Fidelity are among the brokers selling shares. [25:51] CNBC says the top price values Anthropic at up to $1,400,000,000,000. [26:00] And that's your TAWS top five. [26:02] Which one was your play of the week? [26:04] Tell us in the comments. [26:05] We love to hear it. [26:07] Way to go. [26:08] That's the show. [26:09] Today had one thread. [26:11] Who's checking the work? [26:12] Lawmakers question whether Washington properly oversaw Nvidia's chip sales to China. [26:17] A court said it couldn't weigh whether AI wrote the facts in a police report. [26:21] A former AI, OpenAI researcher, checked Chinese AI models for a banned Nvidia chip's fingerprint. [26:28] That's Will DePue. [26:30] You should go follow him on x. [26:32] And four nonprofits asked a judge to stop five AI companies new models. [26:38] Ron didn't have much to say today. [26:42] Maybe he'll be back next week. [26:43] Thanks for all of the comments. [26:45] Keep the one word answers coming. [26:47] Got a surprise meeting invite, make sure that you check who the sender is. [26:51] And always remember, have a great weekend. [26:54] We're here on Monday, again, for AI News RealContext, and I'm your human in the loop. [27:01] I'm Austin Wilson, and we'll see you on Monday. [27:04] Let's go. [31:03] Why is Nvidia's mess now Howard Lutnick's problem? [31:06] Reuters says the White House is weighing pushing out the commerce secretary. [31:12] So why hasn't a new Nvidia CEO enough? [31:20] Why is Nvidia's mess now Howard Lutnick's problem? [31:23] Reuters says the White House is weighing pushing out the commerce secretary. [31:27] So why isn't a new Nvidia CEO enough? [31:35] Did you know AI freed a 100 mile per hour driver? [31:39] ABC fifteen says an AI assisted report got him released, but the report wasn't short on details. [31:48] How'd someone spot banned Nvidia chips in Chinese AI? [31:53] A former OpenAI researcher says six Chinese models carry the band chip's fingerprint, but publishing it could backfire. [32:04] Could a judge freeze ChatGPT's next upgrade? [32:08] Public Citizen and other nonprofits sued OpenAI and Google to halt new models, but slowing AI down has its own lawyers. [32:20] Is everyone on your Zoom call really there? [32:22] FBI says North Korea's fake calls stole over $400,000,000 in crypto, but whose fur faces were they? [32:32] Is Google searches free AI the cheap version? [32:37] Google's putting its fast, cheaper Gemini four hydrant... [32:41] Hydrogen model in search worldwide, but one score tells a different story.