Anthropic Releases Opus 4.6, Software Stocks Tumble Again - DTNS 5201

0:01This is the Daily Tech News for Friday, February 6, 2026. We tell you what you need to know, give you the important context, and help each other understand. Today, Anthropics Claude Opus 4.6 has financial services in its sights. Oh my goodness, it's coming for everything. I'm Jason Howell. I'm Winthrop. Now. Let's start with what you need to know with a big story. All right, let's talk about Anthropics and their new model, Claude Opus 4.6. Now available to users. I've actually played with it a little bit myself. It's designed to be a number of things. But one of the things that is really adept at, and I think they've trained it to be this way, is carrying out financial research.

0:45Also other work-related functions, which we can talk about. But Anthropic is positioning it as an analyst grade engine that can look through, you know, company data, regulatory filings, market information, all that stuff and produce detailed financial research on the other side of it. Now, the model can also manage other tasks like spreadsheet creation, creating presentations. Yes, software development. News of this release immediately impacted financial services on Wall Street. There were some drops of up to 10 % following the news on certain financial companies. And, you know, that might look at this and go, hey, wait a minute, are they coming for our lunch?

1:30Anthropic also announced that Opus 4.6 operates with something called agent teams, which is essentially a way of splitting larger tasks into segmented jobs for parallel work that's kind of coordinated amongst those agents to do that work. That's available in a research preview for API users, also for subscribers. The model has a longer context window, actually offering 1 million tokens of context. We've seen this from Google before with Gemini. Now we've got that here. Axios also reported that Claude Opus 4.6 found more than 500 previously unknown high severity vulnerabilities in common libraries with little to no prompting to do so, at least according to Anthropics.

2:18So Claude Opus 4.6, they call it Opus because it's like the best. It's Mr. Holland's Opus of AI. That was a nice tie back to like the little musical opening we had. Yeah. It's really interesting. I am trying to like be a little open-minded because I definitely am more the more negative people, but obviously the prospect of many areas of things that normally would have that were more manually automated or done through software is definitely, there's definitely some speculation obviously from the market that these could be the tool that kind of overtake the current like ecosystem of products. I mean, I definitely have been using it for work lately.

2:59And I think one thing that I'm trying to be more open for is research. And like, for example, at my job, we have so many docs, like docs upon docs and decisions upon decisions. And yeah, as a human being, it is incredibly difficult. Just me and my little dinky little job with like a couple dozen documents to keep things together. So obviously for something like financial research, I think there was one that they even talked about legal. Am I wrong about that? There was like a legal. there's like even like rumors of like legal like any anything that kind of is like a broad summation or research thing definitely seems like a place where this could be very very good um so i guess we'll see i i still wonder of course whether you know i guess how much how much human review is required uh still and and always and i think the thing that really bugs me is that i think in theory i think we're all okay with human review i mean even if like say you had a bunch of like research assistants or interns playing this together, presumably you'd want someone senior to look over things with like a more experienced eye.

4:01And I think that we understand that for AIs, but I'm a little worried, especially given somehow the incentives are, do we have the motivation to continue human review when, you know, granted these things are so good at doing things so quickly. At some point, are we going to lose that human review and are we in a state where we could get by with that human review? I don't think so, but I don't know. I'm still kind of nervous. Yeah. I mean, they're good at doing things quickly. Are they doing things correctly quickly? Yeah. And I think just as a user, I've noticed this in my own use, really getting into the weeds with these models over the last couple of years, the more you use them, the more you kind of get a little comfortable with the breadth and the speed of that information.

4:49and you can get lazy. You can just be like, oh, well, there it is. I trust it. It's probably fine. Right. Like, do I have 10 minutes to comb through this thing? Like, you know, I've made that mistake and then I've paid the price on the other end. And so, you know, that's a that's a lesson learned and you have to be reminded of that. But when we rely on these systems to do more and more, you know, financial analysis and everything and the results come so seemingly easy to us compared to what they would have been before. You know, on one hand, it's like, okay, well, the results are there. They came easier.

5:26Let's then spend the time that we would have spent creating, checking to make sure. And that just kind of falls on the hand of the user. But I think there is the, I don't know, there is the kind of tendency for people to kind of sidestep that a little bit and be like, eh, it's going to be fine. Or I'll spend 10 minutes and really it required an hour or whatever the case may be, no matter what, it's a time savings. You just, that doesn't mean you're completely absolved, you know? Yeah. That's like, I think my main concern these days is that because these things can do things so quickly that it seems like, okay, we could like 15X, a thousand X the work.

6:04But yeah, I don't know. But it does seem like, I really like the idea that, and I guess I don't quite understand all the context around the Axios report of 500 previously unknown high severity vulnerabilities found in common libraries. That I do think is pretty interesting just because I feel like that's such a good use of AI because it's good at pattern matching. It's good at finding certain things. So I think actually that's pretty fascinating and interesting where you have a very concrete like workflow or very concrete task. And especially with this new, with these agent teams where you can kind of segment it and coordinate it and parallelize it.

6:33That does seem super, super interesting. Obviously they would probably want, I mean, obviously again, it'd be great if a human check this, but I do think that it's weird because there's so many places where this thing shines so well. Um, and it's more like, can, do we know how to smartly segment the parts where this is super, super useful from the parts where it might bite us in the butt? Um, which is honestly a little, little frightening for me as someone who doesn't know anything about financial, um, data and financial research, um, whether, whether this is doing, doing you a favor or doing you a disservice.

7:07Um, but yeah, I mean, Yeah, I'm really curious. It's interesting. It is branching out. I don't know. Would you use it for financial research? I mean, I would test. Sure, I would test with it and kind of see what it comes up with. I mean, you know, money, numbers. These are things that like generally in my life I don't enjoy very much. I've learned to live with it and I've learned to learn it and to find an understanding around budgeting and everything like that. So there is certainly a tendency inside of me that is like, oh, would this make aspects of that easier for me or whatever? But like you said, things like finance, things like legal, these aren't just – it doesn't matter if it gets it sort of wrong.

7:55That was a slight inconvenience. These are more than inconvenience. When you talk about hallucination, it's mandatory in these domains that they get it right. And I think that's really where the jury is still out on this stuff is, you know, you talked about legal. That was from a story earlier this week, Claude Cowork and a feature plug in, which made it very good at doing some of the more mundane legal work. And then, of course, this stock market responded negatively for any of the legal software out there. Like there's a lot of displacement concern around these things. But I think also very much it's about trust because financial, because legal avenues require trust.

8:34it's really juries still out on whether these systems can do this effectively enough when you compare it to human attention. And that's why, you know, I continue to believe personally that the human is never removed from the situation. We always got to be connected. Please keep us connected. Yeah, a hundred percent. I mean, like, I guess one more quick anecdote from this week, for me, I've definitely been using AI to try to help me, like, say, generate test cases for software development. Because it's pretty good about saying, hey, like you should test this thing, like very high level, right? Like, oh, here's a, like, think about this case or think about that case and think about this case.

9:08Now, when I generate the actual code for the test, but I do think that the idea of this high level again, and it's like out there finding patterns, right? Of like, okay, someone else did something like this. These are the test cases that they looked into, which is valuable because that does take time and that can be something that is overlooked. So I definitely applaud it for that. On the other hand, I was talking to one of my very good friends who's a software developer and he just, you know, in all good faith decided to try to generate some tests. Um, just, just let an LM during some tests. And may I say the quote was, these are some of the stupidest tests I've ever seen.

9:42Um, so right now I think, I mean, and it's not very novel. No. And, and the thing is, is that it's also dependent on the platform, what you're doing and even the specific model. So I think, yeah, the trust right now is so interesting. And I guess as like a regular person, I really wonder, should I worry of like the stock market, you know, the stock market trending, like the stock market trends and like the drops in like, you know, in legal software and financial software, not necessarily my for myself for my job, but I wonder what that says, like as a, just a person who exists in this space and, and I guess hires people that uses that kind of software.

10:17I don't, I just, sometimes I'm not quite sure how to interpret all of this, like, you know, movement and, and, and what, and like kind of speculative aspect of the stock market. So I don't know, is confusing. But we hope we can provide a little more clarity for you, the listener. And by the way, DTS is made possible by you, the listener. We want to take a little bit of time to thank HiTechOki, Chris Zaragoza, and Mike Akins. Thank you. I've got a couple of drumsticks here to give you a little drum off as my celebration. All right. Thank you, guys. All right. There's more we need to know today.

10:51Let's get to the briefs. All right. Well, European regulators say TikTok's core features like infinite scroll, Autoplay and its recommendation algorithm all contribute to an, quote, addictive design system that has detrimental impacts on users' health. In particular, they say these systems encourage compulsive behavior on the part of their youngest users and vulnerable adults. The EU's preliminary findings signal that TikTok might have to redesign the core functionality of the app or face fines of up to 6 % of global revenue. So no timeline was given as to when the final decision in the case will be made.

11:28And TikTok now has a chance to formally respond to the allegations. And TikTok has said that it plans to defend itself against these findings, quote, through every means available. Which because, I mean, everybody knows this. At least it's the thing that's repeated time and time again about TikTok. Their algorithm is so good. And, you know, like they it's not like they're alone in creating an algorithm around their platform. They're certainly not alone in creating autoplay and integrating that or infinite scroll even. And so I think that's what's kind of interesting to me about this is that this is not unique to them.

12:07They're just historically seen as very good at how they implemented them. So good that using these techniques, you know, has the EU regulators concerned, you know, is this going to apply then if this kind of, you know, ends up coming to fruition as it sounds like it will. Will this then at some point apply to others since, you know, infinite scroll is not a new thing. It's in Instagram. It's in all the apps, basically all the social apps. So I wonder what the kind of wider impact could potentially be. Yeah, I'm with you. It's not a novel thing. We've had these things, unfortunately, like fortunately or financially, depending on your perspective forever.

12:48May I say infinite scroll is actually quite difficult to implement. So I feel like, oh man, I bet there's a bunch of people that worked at social media companies that wish this case came up earlier in case they didn't have to implement it. But I mean, I don't know, like I don't know because it's such a core thing. And I feel like, again, this is like something that I think the EU is very, I laud the EU always on their kind of proactiveness for trying to address these particular, these very nuanced and very difficult aspects of technology that have obviously had huge impacts on our, so like on society and how us and especially, yeah, our, our young folks, our kids, you know, kind of interact with media and each other.

13:25I don't, I don't, I don't even know where they're going to go with this though, because it's like the entire central part of the app. So what do you, it's like, what do you, it's almost like they're saying TikTok, don't be TikTok. You know what I'm saying? Yeah, totally. I don't, yeah, I don't disagree, but it's also like, that's pretty, that's a big lift in my opinion. I don't know. Yeah. I'll be really curious what that actually means as far as the impact for the app, if it ends up going through. But that is, I guess, the question, right? Like it has not been formalized. So who knows? Maybe that will change.

13:57And then of course, if it's changed in the EU, will that have any impact on any of the other products outside of the EU? Probably not, especially now that at least here in the US, Yes, we have our own version of TikTok to deal with. So there you go. Amazon posted its earnings report showing its fastest growth in 13 quarters, 24 percent. Still beating expectations, missing slightly on earnings. The company is directing$200 billion. And this was really the big story to come out of this. $200 billion of capital expenditures at its data centers, custom ships, networking and satellites. CEO Andy Jassy says the bulk will go to AWS and AI spending, of course, related to that.

14:43Wall Street responded to this with a 10 % drop because that amount was$50 billion more than the expectations for that. Amazon says that AI adoption is moving customers to the cloud very rapidly. And Amazon is shifting its resources to make room for, quote, an extraordinarily unusual opportunity as it does so. So it's basically saying, well, look, everybody's going to the cloud. It's all driven by AI. We've got to make sure to ramp up. So we got to jack that, you know, that capital expenditure amount up to$200 billion to cover this. And I don't know. I think that's stoking some, you know, some renewed, maybe they never went away, but fears around, you know, bubble.

15:30Like again, once again, is this a bubble that is, you know, slowly but surely building up and this is just yet another data point on that path? I don't know that I believe that, but that's what a lot of people are reporting. Hmm. Well, Spotify is tightening access to its developer mode APIs, requiring all developers to have a premium subscription, cutting each app's sandbox to five testers and deprecating some of the more popular endpoints like new releases, artist top tracks and bulk metadata requests. Spotify is framing this as a response to higher risk exposure thanks to automation and AI, stressing that developer mode is only a sandbox for learning and for personal projects and not meant to build or scale a business.

16:14If developers want to reach more than five users, they must apply for extended quotas that some critics argue favor only large companies. that includes a requirement that they be a registered business, have 250 ,000 monthly active users, and coverage in Spotify's key markets. Well, Wynn, you're a developer. I don't know how many times we say that when you're on. No, no, but it's true. That's what I'm here for. Yeah, yeah. So, I mean, when I saw this story, I was like, you know what? This probably isn't the biggest story floating around, but this is an interesting conversation to have. What this reminded me of was kind of what Reddit went through not too long ago.

16:53I don't know if that came up for you. But as a developer, like, what do you think about this? Kind of the tightening of the ship? And is this headed in the same direction as Reddit? I 100 % thought of Reddit. And obviously, they got a lot of blowback, which I did. Do I remember that correctly? They kind of just said, whatever, whatever. I'll do what I want kind of thing. I forget. I'll do what I want. but they did make it almost like they did make it prohibitively expensive for just regular people to use their APIs so I mean it's their API so of course they have the right to charge whatever they want for and certainly especially with Spotify being you know want like a it's Spotify like it is Spotify so of course the the I don't think that it's so I I personally I'm not really someone who works on APIs and understands kind of like the metrics and kind of like the usage of it very well.

17:45But it does seem reasonable to understand that nowadays with automation AI, it probably would, that's not, that's probably not a non-issue. I do think that this is prohibitively, this definitely feels like basically excluding anyone who is not a large company. I mean, you could probably get me behind, you could probably make me accept with a sour face deprecating some of the more popular endpoints if there's a good reason like bulk metadata requests like maybe like to reduce traffic and to reduce their costs sure but i could probably accept that but when it comes to extending quotas um the registered business and 250 000 monthly 250 000 monthly active users is a lot a lot is huge that is prohibited like even like the most successful mid-grade apps, I bet you.

18:40I bet even money, but that's not even. As a guess, I agree. That's what I was thinking. That's just huge. And coverage in Spotify's key markets, that's just, I mean, it's like they're saying it without saying it, that we really don't give a P, give an S, whatever the abbreviation should be there, really about people who are learning in personal projects, which again is their product, but it feels stinky. I mean, I think that, I don't know. I mean, they're not saying that you can't. They're just saying that they're just kind of lowering the limits and, you know, maybe just kind of pressing back on the fact that like, hey, this wasn't meant to build a business anyways.

19:18This was really just meant to like play around with it. Maybe you capture some ideas out of it and that directs, you know, your intention and your efforts down the line. Yeah. And that's, I mean, that's very fair. If it's for learning, then you're fine. And I suppose Spotify is one of those brands where they're not too keen. I mean, I suppose like Reddit too, they're not too keen on third-party clients. Anyway, they really, really understandably, I mean, I guess somewhat understandably, but still not happy about it that they really want you to use their Prime app, which it's their API. Not crazy about this, but that's how it goes.

19:52That's how it goes. Waymo is using Google DeepMind's Genie 3 model to generate realistic virtual environments for its autonomous cars to train on edge case driving scenarios. The kinds of scenarios that are more difficult to encounter safely in real world testing or even at all. You know, it might be so edge case that just never happens. So how do you train for it? We discussed Genie 3, I think it was last week on the show. This is Google's world model interactive AI system that creates 3D worlds based on text and image prompts. They're playable for up to 60 seconds at a time. I find the whole idea really fascinating, personally.

20:30Waymo is using the system to generate synthetic driving and depth perception data, which is the same kind of data that would be captured and processed by vehicle cameras and LiDAR sensors in the vehicle. The system can ingest also dash cam footage that really exists and turn it into a synthetic world for training. I just think this is fantastic. Honestly, like, you know, the greater minds than any of us, you know, like Jan LeCun, who is constantly talking about the need for more development and world models to really kind of train these systems in a more kind of comprehensive way that adheres to, you know, the humanity of this world that we live in.

21:13I think this is using a system that is designed to generate world models for some sort of use that it's easy to kind of fall back on. Oh, well, it's a gaming development tool and certainly it's going to be used for that. But this is one of the really strong, in my opinion, tangible, useful benefits of developing world model systems like Genie 3. I think this is awesome. No, it's so cool to be able to generate all these test cases in a way faster way that a human being could do it and then be able to, yeah, find those edge cases like I mentioned earlier. Yeah, that's cool. It's just cool, period.

21:48Mm-hmm. Yeah. And I think another interesting point on this, and then we can move on because I know we're going a little long, is just the synthetic data nature. Synthetic data has always been kind of confusing to me because it's like, well, you're just creating data because there's a shortage of actual data for these systems to train on. How do you know that data is any good? And here, I think the system is generating enough accuracy in the synthetic data potentially to help these systems know, you know, what to do in a scenario that would be impossible to replicate. I just think it's really cool.

22:24So, yeah. Love it. All right. Well, do you guys want honest reviews from people who actually bought and live with tech? Well, you need Live With It. Sarah Lane hosts a weekly look at tech that she and others have spent months living with because they needed it. This week, our editor Hammond reviews The Tour Box, an accessory that really speeds up his editing. And you can listen to Live With It wherever fine podcasts are traded or watch at youtube.com slash daily tech news show.

22:53And now some quick headlines that are good to know might make you look really smart at the dinner table tonight. All right. Well, Apple is pulling back on plans for its standalone AI health coach service, Mulberry, and instead moving some of its features into the existing health app over time. Okay. OpenAI released GPT 5.3 Codex, its fastest agentic coding model. It's designed to act like an interactive coworker, building, testing, shipping software, other computer tasks, and all of it with a little more autonomy than it had in previous versions. All right. Well, Reddit CEO Steve Huffman says the company will soon roll out a bot verification and labeling system to protect users' trust and keep discussions authentically human.

23:38On the flip side, there's Rent a Human, a new service that enables AI bots to pay anonymous humans in crypto to be their physical hands for tasks in the real world. This is gig economy for humans by robots or so it promises. And thanks to Echo of Oppenheimer in the subreddit for this unnerving glimpse into the future. I feel like we just got into like a Blade Runner or what's the gentleman who wrote, sorry, the book of the G.O. Like 1984? Yeah. Yeah. Thank you. This could very easily have been like one of our longer discussion stories, to be honest. Maybe it should have been up there. That's, I would say, I want to say that's amazing just as a concept.

24:22I'm not going to, that doesn't apply to the actual. Yeah. Oh boy. What does it portend of the future? Yeah. All right. Well, NASA will let Crew-12 and Artemis II astronauts carry smartphones to capture and share moments from orbit after qualifying the devices and relaxing long-standing rules that prohibited their use during space missions. Google's expanding its Apple-compatible QuickShare AirDrop support from the Pixel 10, where it's been kind of isolated for now or exclusive, let's say, to more Android devices, including those running Snapdragon chips, coming sometime this year. and finally meta is spinning out vibes into its own app following in the footsteps of open ai soar app the new app is currently available but we don't have any word on when it will be publicly available we end every episode of dtns with some shared perspectives and today anthony has a question anthony writes i love the show i was listening to the show on last friday and heard the complaining about the windows 11 security requirements i've heard similar complaints around on the internet, but never any discussion of the merits of the security requirements.

25:27There is always a trade-off with usability and security. And I'm wondering what security professionals think of the trade-offs Windows 11 is making. Is the requirement for Windows 11 just security fluff that isn't really worth the upgrades people have to make? Or are they painful but needed changes to the ecosystem for future security? Thank you all for the great show. 10-year listener, Anthony. Yeah, been with DGNS for the long haul. I think that's an important thing to point out, right? Like, you know, I feel like I have to make this statement to my wife when two-factor authentication comes up.

26:04She hates two-factor authentication. I'm always like, look, it's inconvenient. Yes, but, you know, that's how you are more secure is sometimes you have to trade off the convenience for security. Tom did include some notes here because I am not a security, you know, I don't cut my teeth as a security professional in the world necessarily. But Tom put in a few notes here. Microsoft argues that TPM 2.0, which is required for Windows 11, is 52 % more resistant to kernel mode crashes, offers better protection against modern complex malware, particularly at the firmware level. So you got that. There's others that are arguing that older hardware can be capable of supporting the same protections and Microsoft should allow it.

26:49And then another argument is that by restricting Windows 11 installation actually encourages people to stay on older, less secure Windows 10. That's always a given, right? When it comes to like, you should be up to date. You should be on the latest because it's just up to date from a security profile perspective. Yeah, 100 percent. But really good point, Anthony. Always that trade off. Always, always. All right. Well, what are you thinking about? Like, if you have some insight into a story, we would love to hear it. Share it with us. Feedback at DailyTechNewsShow.com. Yeah. Thank you, Anthony, for contributing to today's show.

27:22And thank you. That is you, the rest of you for being along for Daily Tech News Show. You can keep us in business by becoming a patron at Patreon.com slash DTMS.

27:36This week's episodes of Daily Tech News Show were created by the following people. Host, producer, writer, Tom Merritt. Host, writer, Jason Howell. Co-host, Sarah Lane. Co-host, Rob Dunwood. Co-host, Winthred Dow. Producer, Anthony Lemos. Producer, Roger Chang. Editor, Hammond Chamberlain. Editor, Victor Bognat. Contributing producers, Kevin Tech, Noel Cow, and Brandon Richards. Science correspondent, Dr. Nikki Ackermans. Social media producer and moderator, Zoe Detterning. Our mods, Beatmaster, Wscottis1, BioCow, Captain Kipper, Steve Guadarama, Paul Reese, Matthew J. Stevens, a.k.a. Gadget Virtuoso, and J.D.

28:13Galloway. Mod and video hosting by Dan Christensen. Music provided by Martin Bell and Dan Luters. Art by Len Peralta. Acast ad support from Tatiana Matias. Patreon support from Bobby Wagner. And thanks to all our patrons who make this show possible.

28:36the dtns family of podcasts helping each other understand diamond club hopes you have enjoyed this broker