# OpenAI and Anthropic's Enigma Breakthrough

An episode of [AI Chat: AI News on Claude, ChatGPT, Gemini, Grok, Perplexity](https://invisiblenetwork.ai/podcast/1669990482), 29 Sep 2026, 17 min.

Audio: https://content.rss.com/episodes/365073/3194420/ai-chat-podcast/2026_09_29_18_30_57_ae1fc9ba-1ebc-4195-8368-f4ccdd5459c6.mp3

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! In this episode, we feature the successful decryption of WWII Enigma messages using AI technologies from OpenAI and Anthropic. Discover how these advancements enhance our understanding of historical intelligence.

## Transcript

[0:02] Welcome to the AI Chat Podcast. I'm your host,
[0:04] Jaden Schaefer. Every day I cover cutting -edge
[0:07] AI news and talk with the leaders behind it,
[0:09] breaking down what it means for your life and
[0:11] business. Welcome to the podcast. Today we have
[0:17] the United States government that has launched
[0:20] America .gov, a new AI agent for every government
[0:24] website. We have NVIDIA, who's opened something
[0:27] called OpenShell. and Century and it is going
[0:30] to contain rogue AI agents. We also have OpenAI's
[0:34] Astra and also Anthropix's Claude Opus 5 that
[0:38] have cracked some unsolved Enigma messages from
[0:41] World War II. And we have Anthropix founders
[0:44] that are looking to get a 50 .1 % voting control
[0:48] of Anthropix ahead of the IPO. We also have an
[0:52] appeals court that is upholding the Pentagon
[0:53] supply chain risk ban on Anthropix. And we have
[0:57] different Tesla workers that are resisting training
[1:00] Optimus robots to replace them, according to
[1:03] some allegations. And a story about UpGuard that
[1:06] found that 16 ,000 Supabase databases are leaking
[1:10] personal data to the open web. All of that and
[1:13] more on the podcast today. Before we get into
[1:15] the show, I wanted to mention, if you've ever
[1:17] wanted to use all of the different AI models
[1:20] in one place, I'd love for you to check out my
[1:22] startup, which is called AIbox .ai. We give you
[1:25] access to over 90 different AI models in one
[1:28] playground. So you can sit there and you can
[1:30] talk to Claude in the same chat thread as ChatGPT
[1:33] and as Gemini and as Grok. You can then get different
[1:37] image generators and audio generators and video
[1:39] generators all in the same thread. So if you
[1:41] want to get access to all of the different AI
[1:43] models for $8 .99 a month, you can test everything
[1:46] new that comes out. Sometimes you might just
[1:48] need. One video generated for a project, but
[1:51] you don't need to go and spend, you know, $70
[1:54] a month on a high subscription tier of Higgs
[1:56] field. If this is the case, I'd love for you
[1:58] to go check out the AI box platform at AI box
[2:01] .ai. And also we have a connector that plugs
[2:03] it into Claude. So if you want Claude to be able
[2:05] to generate images and video right inside of
[2:07] Claude, go check out the AI box MCP connector.
[2:11] All right, let's get into the episode today.
[2:12] So, I mean, this is a big thing that has been
[2:15] talked about all over X today, which is that
[2:17] America .gov has just launched and it's basically,
[2:21] they're calling it a single front door to the
[2:24] federal government. So instead of opening a whole
[2:26] bunch of different tabs across different agencies
[2:27] and sites trying to figure out which information
[2:30] actually matters to you, you can ask America
[2:32] .gov a question and you get a relevant answer
[2:34] from across 29 ,000 official government websites
[2:38] with the source and what to do next. To be honest,
[2:41] so that was a tweet from Tyler Kim on this, who's
[2:43] working on this. I think that for what it's worth,
[2:46] I mean, everyone's got different political opinions
[2:48] in the United States. However, this is something
[2:51] that I think everyone should get behind in a
[2:53] nonpartisan way, making the government's websites
[2:56] less horribly ugly and confusing. That should
[2:59] be something that everyone should be on board
[3:01] with. So america .gov, I was testing it out this
[3:03] morning. You basically, it's just like a chat.
[3:05] It's a chat thread that you can talk to and it
[3:08] will answer you. Any questions, the big pitch
[3:10] they have on there is that it's going to get
[3:12] the answers only from official government websites
[3:14] and sources. It's free to use and it never has
[3:17] ads and it keeps all of your private information
[3:19] protected. So as I mentioned, 29 ,000 different
[3:23] websites. I hope every government does some version
[3:25] of this. But, you know, I, for one, I'm originally
[3:27] from Canada. I got a green card and I'm living
[3:30] in the United States. And the process for getting
[3:33] a green card and living in the United States
[3:34] was just, I mean, probably in any country, pretty
[3:37] painful. confusing and long and hard and all
[3:41] of the different things. It's doable, but it's
[3:44] not a fun experience, I would say. And same thing
[3:47] with starting businesses, opening LLCs, contacting
[3:50] different departments, getting different tax
[3:53] license stuff, filing taxes, getting passports.
[3:56] I mean, there's a thousand different things that
[3:57] everyone has to do. This should make it all easier.
[4:00] What's cool here is, I mean, right now it's just
[4:01] a chat. chat bubble but they said that coming
[4:04] in 2026 you're going to be able to do a whole
[4:07] bunch of new things one of those being that you're
[4:10] going to be able to actually go and file your
[4:13] passport online including the photo which is
[4:17] pretty wild 2027 you're going to be able to go
[4:21] and book you know like any government campsites
[4:24] i don't know if you've been to government campsites
[4:26] but you have to like it's like a little piece
[4:27] of paper and you got to put like money in a drop
[4:29] box and i mean in the park ranger gets it it's
[4:32] pretty wild so you'll be able to do they're doing
[4:35] campsites for national parks they're doing passports
[4:38] they're doing medication i think there's something
[4:40] called trump rx i don't know too much about it
[4:42] but i know that yeah it's it's you know you'd
[4:45] be able to go get medicine they're going to be
[4:47] doing things with housing and also if you need
[4:50] to like update your name so you know you get
[4:52] married you have to do a legal name change it
[4:54] can update the social security website your mytravel
[4:57] .gov website your cbp your global entry tsa pre
[5:01] -check all that kind of stuff in one spot so
[5:04] anyways I'm really excited when the government
[5:07] is modernizing. I believe this whole project
[5:09] was spearheaded by one of the co -founders of
[5:11] Airbnb, decided to just basically volunteer or
[5:15] work very cheap for the United States government,
[5:17] went in and is redesigning all of the different
[5:19] government sites and implementing AI in there.
[5:22] And so, I mean, honestly, civic duty, nonpartisan,
[5:25] massive kudos and a shout out to this guy for
[5:28] making everyone's life better. So yeah, hope
[5:31] that all governments can do this, but also. There's
[5:34] a lot of companies that have really old, outdated,
[5:36] ugly websites. You know, the insurance industry
[5:38] and a lot of other places. So I think there's
[5:40] a lot of people that could take notes from this.
[5:42] Okay, let's get into what's going on with NVIDIA.
[5:44] They just launched something called Open Shell.
[5:46] And the idea is that it's going to help contain
[5:48] rogue AI agents. They actually partnered with
[5:51] Anthropic on this, but it's basically an open
[5:53] source agent safety platform. And, you know,
[5:56] we have open AI going and hacking, hugging faces
[5:59] and a bunch of other place and a bunch of other
[6:01] things. So there's obviously a lot of drama in
[6:04] this space. And you have people like Anthropic
[6:06] saying, hey, look, we need the government to
[6:08] go regulate everybody and not let any new AI
[6:10] models come out or ban open source, right? There's
[6:12] a lot of different conversations being had. I
[6:14] love NVIDIA's approach. It feels like they're
[6:16] the adults in the room. They've gone and actually
[6:18] created a product, Open Shell, that is addressing
[6:20] this. And like, let's be honest, they're working.
[6:23] in tandem with Anthropic and a bunch of other
[6:25] partners. I think SpaceX is using open safety
[6:28] or open agent safety platform for their cursor
[6:31] agent and their Grok models. Anthropic is building
[6:33] it into Claude. There's a bunch of other people
[6:35] that are working on partnerships with it. OpenAI
[6:37] is part of this effort, but they didn't explain
[6:39] exactly how they were using it. At the end of
[6:42] the day, I think the industry can and should
[6:45] regulate itself. It's definitely capable. That's
[6:47] where all the talent is anyways. I don't think
[6:49] the government has, I mean, you know, we have
[6:52] If the government was the expert on this AI stuff,
[6:54] on the frontier AI technology, they would be
[6:56] putting out, they'd be putting out the technology,
[6:58] but instead asking them to, you know, someone
[7:00] like Anthropic, who's got the smartest, you know,
[7:02] quote unquote, smartest minds and PhDs that are
[7:05] discovering the technology, that are building
[7:06] the frontier, that are making the agents, asking
[7:09] the government to regulate them. I mean, like
[7:11] we mentioned in the last segment, we barely got
[7:13] the government's websites not looking like they're,
[7:15] you know, banks from 1995. So I think the best
[7:19] place to have some of this regulation happening,
[7:21] the people that know the most about it are people
[7:23] in the industry. Having them all work together
[7:26] in a, you know, cross -platform way is fantastic.
[7:30] I think NVIDIA does a great job in this. And
[7:33] yeah, I'm rooting for more of this in the future.
[7:35] Okay, opening eyes Astra and Anthropix Claude
[7:38] Opus 5. have just cracked some unsolved Enigma
[7:41] messages. So there's two crypt analogists that
[7:45] were using LLMs to break wartime codes that hadn't,
[7:48] or basically no humans have been able to crack
[7:50] these for over two decades. I love this concept
[7:53] because especially, you know, obviously in wartimes
[7:56] back in the day, we're sending covert severing
[7:59] messages in World War II, and people were trying
[8:02] to decode what was happening. And there was entire
[8:04] departments, you know, in the government that
[8:06] were... dedicated to this. And yet, all these
[8:09] years later, there was still a bunch that had
[8:11] been completely uncracked. So developer Carter
[8:14] LaFenne was using Astra to decode these that
[8:17] some of these things had been unsolved since
[8:19] 2025. Cryptoanalogist Jack Willis used Anthropix
[8:23] Cloud Opus 5 to break a second unsolved wartime
[8:25] message on September 21. Only seven unbroken
[8:29] Enigma messages remain. There's also one that
[8:32] the plain text is known, but the code is not
[8:34] known. And so you know, for basically there's
[8:38] like the encrypted message gets sent. And when
[8:40] you decrypt it, it's like the Eagle star five,
[8:43] seven walks. So like we decrypted it into the
[8:46] words, but we don't know what those words necessarily
[8:47] mean. So this is, it's kind of interesting when
[8:50] you can get these AI agents because they can
[8:52] look at so many different angles and so many
[8:54] different ways to decrypt some of this stuff.
[8:56] It's really fascinating. Astra built its own
[8:59] Enigma simulator and then it ran. archival research,
[9:02] and it reached a solution in two days on this.
[9:05] So I think this is a fascinating kind of topic
[9:08] for some different research projects. Right now
[9:10] we have the founders of Anthropic that are looking
[9:13] to get a 50 .1 % voting control ahead of their
[9:16] IPO. Dario and six of his co -founders want super
[9:19] voting shares to keep control after Anthropic
[9:21] goes public at evaluation, which right now is
[9:24] getting up to about $1 .5 trillion. I mean, it's
[9:28] interesting because It's Daryl and some of the
[9:30] co -founders that would collectively be getting
[9:32] that. So, I mean, to be fair, none of them are
[9:35] all necessarily doesn't mean they all have to
[9:37] vote together as a block, not saying that they
[9:38] wouldn't. This is similar to what Zuckerberg
[9:41] has with Facebook, but he's one of the few companies
[9:43] that was able to keep super, super majority voting
[9:46] shares. And, you know, some people would argue
[9:48] that's a good thing or bad thing. Some people
[9:49] would say, hey, you know, it's terrible for Zuckerberg
[9:51] because he pursued the metaverse and it just
[9:53] burnt a bunch of money and was useless. But at
[9:55] the end of the day. I mean, that's how he's probably
[9:57] gained control of his company. And he was able
[9:59] to take other big bets like, you know, spending
[10:01] billions of dollars on llama models when it was
[10:03] making no money. It seems like that bet is now
[10:05] paying off. And even the meta -perverse bet,
[10:07] which, you know, I admittedly have criticized
[10:09] a lot over the years, everything that happened
[10:11] with like Oculus, which, you know, I don't see
[10:13] people walking around with Oculus these days,
[10:15] not to say it's not used, but it's just not as
[10:17] big as I think they hoped. All of that technology
[10:19] is going into the meta -ray bands and to a lot
[10:21] of these. glasses technology and augmented reality
[10:24] there. So I think a lot of the money spent there
[10:27] is getting recycled and reusing some of these
[10:29] newer products and kind of evolution of these
[10:31] products, which was probably just too much of
[10:33] a big step back then to make. So, you know, maybe
[10:35] it maybe it wasn't bad. But all that to say,
[10:37] getting the super voting is. This super majority
[10:42] voting shares is something that they're going
[10:43] after. Each of the seven co -founders, including
[10:46] Dario, own about 2 % of Anthropic at this point.
[10:49] Anthropic was valued at $965 billion in May,
[10:55] but they have traded at $1 .5 trillion on the
[10:57] secondary market. So that's basically the employees
[11:00] of the company are being able to sell. There's
[11:02] a bunch of secondary markets where they can sell
[11:03] their shares and people are paying up to a $1
[11:06] .5 trillion valuation for those shares. Founders
[11:09] board seats would grow from two to three, while
[11:12] the long -term benefit trust would still choose
[11:14] most of the board. So Anthropic kind of has an
[11:16] interesting way that they have structured this
[11:19] up. And this proposal comes at the same time
[11:22] they have kind of a bigger quote -unquote wealth
[11:24] commitment. Dario announced in January that the
[11:27] co -founders had pledged to give away 80 % of
[11:29] their wealth, warning at the time that AI -driven
[11:32] wealth concentration would destabilize society.
[11:35] So, I mean, this is only a five -year -old company.
[11:37] They're getting close to, you know, a trillion
[11:39] dollar plus IPO. And they're planning on giving
[11:42] away a lot of their funds. I mean, basically
[11:45] saying that the inequality from AI was going
[11:48] to kind of wreck the world in some ways. So it's
[11:51] interesting to see what's going on here. And
[11:53] it looks like they want to keep most control
[11:55] of the company. You know, some people positively
[11:57] would say so that it doesn't get derailed. And
[11:59] other people would, you know, just say they're
[12:00] trying to keep control. At the end of the day,
[12:02] I think this is pretty standard for anyone starting
[12:03] a company. The appeals court is also upholding
[12:06] the Pentagon supply chain risk ban and on anthropic.
[12:09] There was a two to one DC sort circuit ruling
[12:12] that keeps Claude blocked from the defense department
[12:15] systems and contractors, despite a parallel.
[12:18] san francisco lost for the pentagon so i mean
[12:20] this is going through many different courts in
[12:22] a bunch of different ways in san francisco they
[12:24] lost but in dc they upheld this so i imagine
[12:28] this is something that's going to continue to
[12:30] be litigated the circuit judge gregory katas
[12:33] and nomi rao formed the majority circuit judge
[12:36] karen lecraft henderson dissented on that a san
[12:39] francisco federal judge ruled last month that
[12:41] a parallel dod designation was illegal. So now
[12:45] we kind of have these two different things. Anthropic
[12:47] said that is considering all options, including
[12:49] further review. So I think that's not the end
[12:51] of the road for Anthropic trying to get their
[12:54] AI into the government, but we'll just have to
[12:56] see where that goes. It seems like there's definitely
[12:58] some very frosty relations with the Department
[13:01] of War. Tesla workers right now apparently are
[13:05] resisting training Optimus robots. because these
[13:08] bots are built to replace them, right? If you're
[13:10] working on robots and they're like, hey, we need
[13:12] to build robots that can help work on these robots,
[13:15] maybe you don't want to do that. Tesla right
[13:17] now is aiming for a thousand Optimus robots created
[13:20] every week by the end of the year. Apparently
[13:22] there's a bunch of workers pushing back and there
[13:24] is a number of reasons. I mean, sure, workers
[13:28] are pushing back, but I mean, if your job is
[13:29] to build robots, you probably know what you're
[13:32] in for there. But another part of this is that
[13:34] the hand on the Optimus robots has about 100,
[13:37] plus different parts. And that's also slowing
[13:40] the ramp down quite a bit. So apparently this
[13:42] is all happened in the Freedmont factory in California.
[13:44] They stopped making the Tesla Model S and Model
[13:47] X in May of this year to shift all of the workers
[13:50] and engineers onto creating the Optimus bot.
[13:53] So that was kind of a big swing that Elon took
[13:55] with Tesla, basically saying, look, we have these
[13:58] very popular cars, the Model S, the Model X.
[14:01] These are very popular cars. We're going to completely
[14:02] stop making them and just put all of that bandwidth.
[14:05] into creating the Optimus robot because we see
[14:07] so much potential there. Optimus is very complicated.
[14:11] Hyundai plans to deploy up to 25 ,000 Atlas robots
[14:14] from Boston Dynamics over the next several years.
[14:17] You can imagine Optimus is going to have a very
[14:19] similar demand from different companies. So if
[14:21] Hyundai alone is planning on grabbing 25 ,000,
[14:24] you can imagine many, many companies will be
[14:26] buying even more than that. The FCC has banned
[14:28] new foreign -made humanoid robots, robot dogs,
[14:31] and robot vacuums in July. So considering Optimus
[14:35] is going to be built in the United States, it
[14:37] seems like this is it seems like Elon might just
[14:40] be and a number of other Silicon Valley companies
[14:43] might be in for a huge boost when all of the
[14:46] foreign bots are banned. Right. We're not going
[14:49] to have a whole bunch of robots, humanoid robots
[14:51] coming out of China. I mean, it's been literally
[14:53] foreign made. So that's. really wild take. Now,
[14:57] you can you can see like a lot of the reasons
[14:59] or the reasoning behind a decision like that,
[15:02] right? Like, theoretically, let's say China released
[15:05] like a million humanoid robots into the United
[15:08] States, and they're helping people. And it's
[15:10] basically like DJI drones, but like robots. And
[15:13] then all of a sudden, you know, there's like
[15:14] a backdoor they create and all the robots. turn
[15:17] into an army and go attack a military base. Like
[15:19] you could imagine some terrifying, some terrifying
[15:22] scenarios there. So they definitely want to keep
[15:24] them all created by American companies. And beyond
[15:27] that, I mean, you really have to trust a company
[15:29] making a humanoid robot. People are going to
[15:31] have these in their homes. They're going to have
[15:32] them in businesses. It is definitely a wild time
[15:37] for technology. Okay, UpGuard right now has found
[15:40] that there is 16 ,000 super -based databases
[15:43] right now that are leaking personal data to the
[15:44] open web. I didn't want to, like, do this segment,
[15:47] like, as, like, a fear and doom thing. Like,
[15:48] oh, man, this is, like, vibe coding's bad. I
[15:51] will just say, like, if you are vibe coding things,
[15:53] which I vibe code lots of projects. You got to
[15:56] be careful because a lot of these different database
[15:57] tools, Supabase being one, that's the main, the
[15:59] main database that I use for all of my vibe coding.
[16:02] You got to be careful because you have to regularly
[16:04] conduct security audits, which I mean, at the
[16:07] most basic sense, you can just tell Claude or
[16:09] ChatGP or whatever using vibe code, please conduct
[16:11] a security audit. a security audit. Let me know
[16:13] if there's any issues. I think companies like
[16:15] lovable are doing a good job trying to do that
[16:17] themselves internally and let you know if there's
[16:19] issues and have you fix it. But yeah, it's definitely
[16:22] an issue for all of these people that don't know
[16:24] what they're doing, developing. They're just
[16:26] kind of shouting at the void and it's creating
[16:28] things for them. It's super exciting that you
[16:29] can create things. For example, if you're watching
[16:32] the video of this on screen, I'm showing my website,
[16:34] AI chat daily .com, where I have like all of
[16:36] these articles and a ton of information and a
[16:39] lot of really cool stuff that I've built. But
[16:41] this entire site was vibe coded. However, I did
[16:44] have to run all these security audits on it to
[16:46] make sure that different things in the backend
[16:47] weren't visible, especially when you're dealing
[16:49] with actual customers and people's personal information.
[16:53] So just a PSA on that. But at the end of the
[16:55] day, so much to build, so much to do. Thank you
[16:57] so much for tuning into the podcast. If you enjoyed
[16:59] the episode, make sure to go check out. aibox
[17:02] .ai where you can get access to over 80 different
[17:04] AI models. You can test them all against each
[17:06] other. This is a phenomenal way to save money,
[17:08] consolidate subscriptions, but also get access
[17:10] to all of the latest different AI tools in one
[17:13] place. All right, I'll catch you all in the next
[17:15] episode.

Shows are listed from Apple Podcasts' charts and directory, and Podcast Index where it is used. Every show and episode belongs to its publisher and plays from the publisher's own feed; this site hosts none of them.