本周在Uncanny Valley, Brian Barrett, Leah Feiger，和Zoë Schifer 潜入我们的独家新闻关于一个国防承包商如何将Claude Code 连接到国防部的安全系统我们也在讨论Anthropic的IPO存档中看到的绿色、黄色和红色旗帜，以及为什么这不仅仅是对AI行业的考验
本集提到的文章：
- 国防承包商让其个人AI Agent进入安全五角大楼系统
- Anthropic S-1 存档也是AI末日首页
- 旧金山地产里有什么比现金更重要的呢？Anthropic库存
- 大批志愿人员准备在Polls面对ICE
你可以跟随布赖恩·巴雷特(Brian Barrett)和布鲁斯基(Zoë Schifer)，在[电子邮件保护]给我们写信。
如何倾听
您可以通过本页面的音频播放器随时收听本周的播客节目，但如果你想免费订阅以获得每集节目，则如下：
如果你在iPhone或iPad上，打开名为"Podcasts"的应用程序，或者只是点击这个链接。 也可以下载Overcast或Pocket Casts等应用，我们也在Spotify。
语录
说明：这是一个自动记录，可能包含错误。
欢迎来到WIRED的荒凉谷我是布莱恩・巴雷特，执行编辑。
Zoë Schifer：我是Zoë Schifer，撰稿编辑。
李亚飞：我是李亚飞，政治和科学总监。
Zoë Schiffer: (法语). 本周，Anthropic正式备案公开。 公司作为S-1档案的一部分提交的上百页，从惊人的收入和损失，到对AI存在威胁的警告，都发生了很多事。 很明显，我们说的是Anthropic 我们将讨论我们的关键外卖和为什么这个IPO 是一种很高的利害关系，不只是为AI行业，而且诚实地说整个美国经济。
Leah Feiger：我们也给你带来了关于国防承包商如何将Anthropic的Claude 插入一个安全的国防部系统，在政府中 AI 真正最外向的用途之一。
布莱恩·巴雷特：而中期报告就在拐角处，就在几天后。 全国各地的广大志愿者团体已经准备好保护自己的社区，使其免受ICE官员在投票中展示的潜力的影响。 我们将了解这种基层努力是如何反映一个非常不确定的选举周期。
Zoë Schifer：所以在星期一，我们得知Anthropic终于扣下扳机，公开申请了其IPO. 由于一系列原因，人们非常期待这样做。 Anthropic预计将在IPO中筹集比历史上任何一家公司都多的资金，它将成为第一家只专注于建设AI软件的美国大公司去公开。 为了让大家了解我们所说的规模，公司估值从一年多前的1,830亿美元略增至最近一轮的9,650亿美元，有些人预计一旦开始交易，其价值可能接近2万亿美元。 因此，档案基本上给了我们Anthropic业务内部的第一个真实的外观，因为它现在是一个私人公司。 不须一一一一报。 这些数字在两个方向上都相当极端，数十亿美元的收入甚至数十亿美元的损失，以及公司已经承诺要计算的数千亿美元。 因为在这份文件里还有一吨可以拆开的，我想把它分为绿旗，黄旗，和红旗是很有趣的，我们从档案中看到了。 你们觉得怎么样？
布莱恩·巴雷特：是的。
利亚·费格：我对这里有多少钱感到惊讶。 这是，是的。
什么旗帜，莉亚？什么旗帜？
利亚·费格：我不知道。
Zoë Schiffer：为节约起见为绿色。
布莱恩・巴雷特：太好了。 绿旗。
黎飞杰：为了经济，绿了。 这是这么多钱，伙计们。 但现在我唯一能想到的是，文章说WIRED 几个月前运行关于人们如何在海湾地区卖房子，我还记得当时我说，“我们离这有多近？” Zoe 称，"哦，他们会赚这么多钱"
布莱恩·巴雷特(Brian Barrett)：这公平吗，Zoë，我觉得有财务在那里，是的，他们正在失去钱交接拳头，至少，这感觉像他们把很多关注平息，不是所有的，但这是一个真正的业务。 这是一个正在成长的企业。 它在做所有的事情你希望一个企业做。
Zoë Schifer：是的，我认为绝对如此。 因此其收入在2026年上半年突破了163亿美元，从去年同期增长超过14折。 我认为我们必须回到2022年，并记住当ChatGPT出现在现场时，甚至连AI会有一个真正的企业应用程序都看不出来。 这是一个消费者的应用软件，没有任何商业理由。 直到很多年后Anthropic出现在现场，它才推出Claude Code，我们开始真正看到，哦，不，有一个可行的企业，并且实际上大部分的钱会来自其他企业使用他们的编码产品。 我想这也是达里奥和萨姆之间的关键区别与Anthropic首席执行官达里奥·阿莫代非常相近的人会说他每天醒来，称"企业代码". 这是他关心的这是因为这是为这些企业赚钱的东西。 我认为，有些人认为OpenAI更分散一些，而且到处都是。 我想你可以从数字中看出这一点。
布莱恩·巴雷特(Brian Barrett)：所以他们靠拢了。 他们拥有Claude code，巨大的成功，企业代码，巨大的成功。 我要说，在您所说的收入中，有一个黄旗，令人印象深刻的收入超过600亿美元，其中近一半来自三个技术公司云集市场。 因此，他们相当严重地依赖少数公司，它们也是竞争对手。 我认为Anthropic并没有把它当作它的主要风险因素之一。 我们可以在一秒钟内谈论风险因素，因为那是我最喜欢的部分，但它确实感觉像，护城河在哪里？是否有足够多的证据表明他们依赖少数公司也正在开发自己的AI，其他AI产品越来越好，开源产品，开放重量产品在那里。 那面旗帜多黄？
Zoë Schifer：我觉得是黄色的。 一个多种族的工人合作社在商业交易中乱伦并不奇怪。 我对这个消息不感到震惊。 而小比例的花花公子对公司收入产生超规模影响也并不罕见。 我认为Ramp最近公布的另一个数字，Anthropic和OpenAI收入的80%都来自AI支出的前1%. 因此支出最多的公司占了公司绝大部分的收入。 但它是一个脆弱性，因为如果有一个拉回并且这些公司开始支出更少，Anthropic 暴露。 但另一个，我不知道该放哪里我觉得是绿色的。 它们对于IPO所涉及的一些风险是透明的，对于AI的一些存在性风险一般是透明的。 这可能是我最喜欢的整个事情。 “关切是融资。..... 这种足够先进的人工智能系统如果不得到适当的开发和保障，就可能给人类带来灾难性或生存性的风险。”
离奇的东西。
Zoë Schiffer: (法语). 在S-1中使用被动声音的担忧是达里奥的融资，真实如此。
布莱恩·巴雷特(Brian Barrett)：我也会说，在Anthropic内部，这个短语存在风险有三次出现。 它对人类有两次存在风险，一度对社会存在风险。 所以在这种情况下，大概还有人类，他们刚刚放弃了任何社会契约的外貌。
佐厄·席弗：太棒了。
- Zoë，我爱这是一面绿旗给你，事实上他们是那么公开。..
莉亚·费格：她是一个透明的女士。
布莱恩·巴雷特(Brian Barrett)：老实说，他们从一开始就一直喊着说AI可以杀死所有人。 这对他们来说并不新鲜。 吾所想取之为趣，而取之为取之为取之为取之为取之为取之为取之为取之为取之为取之为取之为取之为取之为取之为取之为取之为取之为取之为取之为取之为取之为取之为取之为取之为取之为取取之为取之为取取之为取之为取之为取之为取取之为取之为取取之为取之为取之为取取之为取之为取之为取之为取之为取之为取之为取为取之为取之为取之为取为取之为取之为取之为取之为取之为取之为取之为取之为取之为取为取之为取之为取之为取之为取为取之为取之为取之为取之为取为取之为取之为取之为取之为取为取之取之为取为取为取为取之为取之取之取为取之为取之取之取因此，他们提到的最有可能导致地球上人类生命终结的事物，也是他们引用的主要竞争优势之一。 这就是递归自我改进。 这就是AI模型可以自我改进的地方。 而他们一方面说，如果AI模型在自我改善方面不断提高，并最终可以自我改善到人类甚至不需要看的地步，关闭AI刚刚好转的循环，使AI变得更好，让AI变得更好到我们甚至无法理解AI正在做什么，但天，它很好，而且赚了大钱，这就是AI强大到可能终结世界的情景。 但它也是，Anthropic 指出我们真的很擅长。 我们为此投入了大笔资金。 故能先发。
Zoë Schiffer: (法语). 多么令人高兴？我们中间谁没有影子的一面被影子的一面赋予你的力量所增强呢？
莉亚·费格(Leah Feiger)：要说清楚，WIRED覆盖了这一点，基本上每个人都如此。 当前Anthropic研究员雅各布·克克森辞职而称是因为Anthropic可以带来世界末日时，他特别引用了抑制RSI在他的榜首。 这是整个球赛。
布莱恩·巴雷特：是的。 然后整个行业非常。.. 我认为这也许太极端了，但伯尼·桑德斯(Bernie Sanders)有一个法案在那里，而罗·汉纳(Ro Khanna)有一个单独的法案，说："嘿，让我们只是禁止递归自我改进。 我们只是说，我们不这样做。” 我要说的是，风险因素是计划书中最长的部分。 应该是94页吧因此，他们可以只拥有称，"我们有一个阴影的一面并拯救了所有人。
我本该理解的
布莱恩·巴雷特：很多麻烦".
Leah Feiger：我们可以谈谈我还没有得到一个好答案的一面红旗吗？我有点需要你跟我解释一下，就像我是个孩子一样？计算账单，所有这些技术花费这么多钱。 我个人并不认为我理解极限所在，你如何没有花费足够的钱来使用你的全部资金，也花费足够的钱来不把自己从商，永远结束你的技术。 或许我并不理解它的整体计算，或它来自何处，但这一切加在一起，这实在是太多了。 告诉我
如此之多。 我不会假装我明白了我们所说的规模是完全前所未有的。 当你开始谈论这些东西的时候，你会得到很多铁路类比，但我会说，一个关键区别是，与铁路公司不同的是，铁路公司运行的边际略微小，这些公司的资金仍然非常充足，很富有。 所以是的，他们投入了一吨的计算，他们期望需要这个计算。 迄今情况如此，它们期望继续上升。 我个人认为这是一个可行的论点，但当然，反面是像，嗯，如果增长开始严重的sl
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
This week on Uncanny Valley, Brian Barrett, Leah Feiger, and Zoë Schiffer dive into our scoop on how a defense contractor connected Claude Code to a secure Department of Defense system. We also discuss the green, yellow, and red flags we see in Anthropic’s IPO filing—and why it’s a test not just for the AI industry, but for the entire US economy—and how volunteers nationwide are preparing for ICE agents potentially showing up at polling sites in the midterms.
Articles mentioned in this episode:
- A Defense Contractor Gave Their Personal AI Agent Access to a Secure Pentagon System
- The Anthropic S-1 Filing Is Also an AI Doomsday Primer
- What’s Worth More Than Cash in San Francisco Real Estate? Anthropic Stock
- Massive Groups of Volunteers Are Ready to Face Down ICE at the Polls
You can follow Brian Barrett on Bluesky at @brbarrett, Zoë Schiffer on Bluesky at @zoeschiffer, and Leah Feiger on Bluesky at @leahfeiger. Write to us at [email protected].
How to Listen
You can always listen to this week's podcast through the audio player on this page, but if you want to subscribe for free to get every episode, here's how:
If you're on an iPhone or iPad, open the app called Podcasts, or just tap this link. You can also download an app like Overcast or Pocket Casts and search for “uncanny valley.” We’re on Spotify too.
Transcript
Note: This is an automated transcript, which may contain errors.
Brian Barrett: Welcome to the WIRED's Uncanny Valley. I'm Brian Barrett, executive editor.
Zoë Schiffer: I'm Zoë Schiffer, contributing editor.
Leah Feiger: And I'm Leah Feiger, director of politics and science.
Zoë Schiffer: This week, Anthropic has officially filed to go public. There is a lot going on in the hundreds of pages that the company submitted as part of its S-1 filing, from staggering earnings and losses, to warnings of AI's existential threat. Obviously, we're talking about Anthropic here. We'll discuss our key takeaways and why this IPO is kind of high stakes, not just for the AI industry, but honestly the entire US economy.
Leah Feiger: We're also bringing you a scoop about how a defense contractor plugged Anthropic's Claude into a secure Department of Defense system, in what's really one of the most outlandish uses of AI in the government we've seen so far.
Brian Barrett: And the midterms are right around the corner, just a few days away. Massive groups of volunteers all over the country are already getting ready to protect their communities from the potential that ICE officers show up at the polls. We'll get into how grassroots efforts like these are a reflection of a deeply uncertain election cycle.
Zoë Schiffer: So on Monday, we learned that Anthropic finally pulled the trigger and publicly filed for its IPO. This has been highly anticipated for a bunch of reasons. Anthropic is expected to raise more money in its IPO than any company in history, and it would be the first major US company focused solely on building AI software to go public. To give you a sense of the scale that we're talking about, the company's valuation went from $183 billion a little over a year ago to $965 billion in its most recent round, and some expect that it could be worth close to $2 trillion once it starts trading. So the filing basically gives us the first real look inside Anthropic's business because it's a private company right now. It doesn't have to report a bunch of stuff. And the numbers are pretty extreme in both directions, billions of dollars in revenue and even more billions in losses, and hundreds of billions that the company has already committed to compute. Because there's a ton to unpack in this paperwork, I thought it would be fun to divide it into green flags, yellow flags, and red flags that we saw from the filing. What do you guys think?
Brian Barrett: Yes.
Leah Feiger: I'm astounded by how much money is at play here. This is, yes.
Brian Barrett: What flag, Leah? What flag is that?
Leah Feiger: I don't know.
Zoë Schiffer: For the sake of the economy, it's green.
Brian Barrett: Great. Green flag.
Leah Feiger: For the sake of the economy, it's green. This is so much money, guys. Sue me, but the only thing I can think of right now is that article that WIRED ran a couple of months ago about how people were selling their homes in the Bay Area with the promise of Anthropic stock post-IPO. And I remember at the time I was like, "How close are we to this?" And Zoe said, "Oh, they're going to make so much money."
Brian Barrett: Is it fair to say, Zoë, I feel like having the financials out there, yes, they're losing money hand over fist, but the financials I think are better than people were expecting maybe. At least, it feels like they put a lot of concerns to rest, not all of them, but this is a real business. This is a business that is growing. It's doing all the things you would want a business to do.
Zoë Schiffer: Yes, I think absolutely. So its revenue hit $16.3 billion in the first half of 2026, and it's grown more than 14 fold from the same period last year. I think we have to go back to 2022 and remember that when ChatGPT came on the scene, it wasn't even apparent that AI would have a real enterprise application. This was a consumer app that didn't have really any business case for it. And it wasn't until Anthropic came on the scene years later, it rolled out Claude Code that we started to really see like, oh no, there's a viable business and actually the majority of the money is going to come from other businesses using their coding product. This is also, I think, a key difference that you hear from insiders between Dario and Sam. People very close to Dario Amodei, the CEO of Anthropic, will say he wakes up every single day and says, "Enterprise code." That is the thing he cares about. And that's because that's the thing that makes money for these businesses. Whereas I think the perception from some is that OpenAI is a little more scattered and all over the place. And I think you can see that in the numbers.
Brian Barrett: So they are leaning in. They have Claude code, huge success, enterprise code, huge success. I will say, to introduce a yellow flag, of that revenue that you talked about, very impressive revenue over $60 billion, nearly half of it came from three tech companies cloud marketplaces. So they are pretty heavily reliant on a small number of companies which are also competitors. It is this sort of tangled nest of reliances in this industry that feels, I think Anthropic didn't put it as one of its major risk factors. We can talk about the risk factors in a second because that was my favorite section, but it does feel like, where's the moat? Is there enough of one given how much they're relying on a handful of companies that are also developing their own AI, other AI products are getting better, open source products, open weight products are out there. How yellow is that flag?
Zoë Schiffer: I think it's yellow. It's not surprising that a polyamorous worker cooperative would be incestuous in its business dealings. I was not shocked by that information. And it's not unusual for a small percentage of spenders to have an out-sized impact on a company's revenue. I think there was another number that Ramp put out recently, 80 percent of revenue for both Anthropic and OpenAI comes from about the top 1 percent of AI spenders. So the companies that are spending the most account for the vast majority of the company's revenues. But it is a vulnerability, because if there's a pullback and those companies start spending less, Anthropic is exposed. But another, I don't know where to put this flag. I think it's green. They were transparent about some of the risks involved in the IPO, some of the existential risks of AI generally. This is probably my favorite part in the whole thing. “Concerns have been raised ... That sufficiently advanced AI systems could, if not developed and safeguarded appropriately, pose catastrophic or existential risks to humanity.”
Leah Feiger: Incredible stuff.
Zoë Schiffer: The use of the passive voice in this S-1, concerns have been raised by Dario to be so for real.
Brian Barrett: Internally by Anthropic, I will say too, the phrase existential risk comes up three times. It's twice existential risk to humanity, once existential risk to society. So in that scenario, presumably there are still humans, they have just abandoned any semblance of a social contract.
Zoë Schiffer: Great.
Brian Barrett: Zoë, I love that this was a green flag for you, the fact that they're so openly-
Leah Feiger: She's a transparent lady.
Brian Barrett: And to be clear, they have been shouting this since they started that AI could kill everybody. This is not new for them. What I thought was interesting, and I'll call it a red flag, red flag for humanity, yellow flag for Anthropic, green flag for me because I love to talk about it. So the same thing that they cite as most likely to cause the end of human life on earth, is the thing that they also cite as one of their primary competitive advantages. And that is recursive self-improvement. So that's where an AI model can improve itself. And they're saying on the one hand, if AI models keep getting better at improving themselves, and can eventually improve themselves to the point where humans don't even need to look at it, closing the loop where AI just gets better, and makes AI better, and makes AI better to the point where we don't even understand what AI is doing anymore, but gosh, it's good and it's making a lot of money, that's the scenario in which AI is powerful enough to maybe end the world. But it's also, Anthropic notes also we're really good at that. We're investing a lot of money in that. That's how we're going to get ahead.
Zoë Schiffer: And how relatable is that? Who among us doesn't have a shadow side that is equally augmented by the power that that shadow side gives you?
Leah Feiger: To be clear, and WIRED covered this, as did basically everyone. When former Anthropic researcher Jacob Coxon resigned and said it was because Anthropic could bring about the end of the world, he specifically cited curbing RSI at the top of his list. This is the whole ballgame.
Brian Barrett: Yes. And then the industry in general is very ... I think this is probably too extreme, but Bernie Sanders has a bill out there, and Ro Khanna has a separate bill that say, "Hey, let's just ban recursive self-improvement. Let's just say we're not doing that." I will say one little note is that the risk factors is the longest section of the prospectus. It's 94 pages, I believe. So they could have just said, "We have a shadow side and saved everybody.
Zoë Schiffer: I would've understood.
Brian Barrett: A lot of trouble."
Leah Feiger: Can we talk about a red flag that I haven't gotten a great answer on yet, and I kind of need you to explain this to me like I'm a toddler? The compute bill, all of this technology costs so much money. I don't personally feel like I have an understanding of where the limit is, how you do not spend enough to use all of your money and also spend enough to not put yourself out of business and end your technology forever. Perhaps I do not understand compute in its entirety or where it comes from, but all of this together, this is just so much. Tell me.
Zoë Schiffer: It is so much. I'm not going to pretend that I understand it. The scale of what we're talking about is, again, totally unprecedented. You get a lot of railroad analogies when you start talking about this stuff, but I will say that one key difference is that unlike the railroad companies, which were operating on slightly thinner margins, these are still very well-funded and rich companies. And so yes, they're investing a ton in compute, and they expect that that compute will be needed. So far it is, and they expect that to continue to rise. I personally think that that is a viable argument, but of course the flip side is like, well, if growth starts to seriously sl
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.