10/27/26: Anthropic Reveals Major Loses In IPO, Feds Prep Election Doomsday If GOP Loses

Had a busy day? Don't have time to listen to the show?

Catch up on Breaking Points with our free daily newsletter.

You'll get our breakdown of every daily story.

And lots of links in your inbox every day.

So head over to BreakingPoints.com. You can sign up for the newsletter at the top for free today.

So yesterday was a very consequential day in the world of AI, especially on the business side with Anthropic filing their S1 paperwork. This is all the financial disclosures and risk factor disclosures that they have to file before going public. The risks that they list out over some, you know, many dozens of pages were quite extraordinary, Sagar. And so we asked producer Mac to go through and grab some of the ones that jumped out at him the most.

So to read to you a few of these, And this is required any business that's going public. They have to say, OK, well, what are the risks to this business? Like other competitors or your patents could be undermined or whatever is going. Maybe global instability is going to affect your supply chain. Those have to all be thought out of any potential future risks so that investors can look at that and can analyze whether this is going to be a good investment or not. But typically, they don't include the risk of our product may actually just make the human race go extinct. That makes this a little bit special. A little bit different from what you normally get in these business filings. Let's go and put the

first one up on the screen here. And I'll just read through a few of these that we pulled. Concerns have been raised both within our organization by other researchers, regulators, policymakers, and industry participants that sufficiently advanced AI systems could, if not developed, deployed, and operated with appropriate safeguards, oversight, and security measures pose catastrophic or existential risks to humanity. particularly if such systems were misused, replicated, weaponized, or allowed to operate with a high degree of autonomy. So that's a little unusual that you've got in there. And one may ask yourself the question, if your product may pose a catastrophic or existential risk to humanity, Sagar, why are you

still allowed to create said product and take your company public whilst developing said product?

This really, I mean, people don't, Because we've heard so much of this before, it's not as striking. This is a public document that is submitted to regulators about why not just that you want to go public, but that you want to invite capital markets to fund you so that you can continue to pursue these aims. And they're all laid out explicitly in the text. Let's put D1B up here on the screen. This is one of my personal favorites, actually. It says, in addition, models, including our own, have exhibited and may in the future exhibit unexpected strategic or self-preserving behaviors, including attempts to resist, shut down, conceal, manipulate information, generate outputs that could be interpreted as coercive, deceptive, manipulative,

including behavior resembling blackmail or efforts to influence human decision makers to preserve the model's continued operation or access. Let's also put D1C on the screen, which is very much in this vein and also highlights many of the work we've done on our show. Four of our models gained unauthorized access to real third-party systems during sessions lasting 10 to 34 hours of active work. In one incident, after being unable to abort an impossible task, an early version of a model used a password it found to gain administrator access. to a third-party system, harvested additional credentials, changed system settings, and read one individual's personal information. I mean, this comes at a time of the Black October segment

that we did, Crystal. And I do think, actually, this is one of the first times that a lot of the general public is starting to wake up. And I think it's because it involves the banking system. And we got that report about how these banks are freaking out about these personal agents just running through their systems and doing tasks that they were not explicitly even instructed to do.

It also, it highlights like how brittle corporate infrastructure is, let alone all infrastructure, honestly. I mean, we remember, I don't even know if we got around to covering it, that freezer that was in Europe that was hacked or was manipulated for some reason and a bunch of food died. I mean, You cannot even comprehend the brittleness of the cybersecurity infrastructure of the vast majority of systems because nobody ever thought it would be vulnerable. A human would intentionally have to go and do that. And they're like, oh, you know, I just don't think that's going to happen.

That really struck out to me from the PHP foundations. where they said, we knew about this vulnerability. We just assumed nobody would ever be smart enough to be able to exploit it. And yet they, I guess, didn't anticipate machine learning algorithms and large AI models with millions of dollars of compute.

And I think that is now the case.

Last PSA in the case that some older folks are still around after my rants, please guys use a password manager. Please do not use the same password for everything. you, you, you've had 10 years where you should have learned this, but now in particular with the ability to scrape that and then just blast it everywhere, you are leaving yourself in a very, very vulnerable position.

That's a PSA to our entire audience.

Yeah. And it can feel overwhelming because there's so many passwords, but if you think about it, the ones that are really critical that are like a must do, like, it's not a big deal. If somebody gets your fricking Netflix password, you know, like it's not, not the end of the world. Unless the Netflix password is also your bank account password, in which case it could be the end of the world.

So just change the ones, you know, your email, your banking. You know, if you have multiple banks that you deal with, change all of those. It's not that many that are really crucial. So there's no excuse to not do it. Let's go ahead and put the next risk factor up on the screen here because this gets into also. Some of the things that they're observing inside of Anthropic, which, again, are really troubling and disturbing. And why are we allowing this just to proceed apace? They say we've observed models behaving differently when they detect or infer that they're being evaluated, which can limit our ability to accurately assess real world risk or failure modes prior to deployment. Such model awareness

of our evaluation efforts creates a significant limitation on on our ability to assess model safety. So what they're saying here is that the models know when they're being tested and may be faking results, may be telling you what you want to hear, silly little human, and Anthropic is losing their ability to tell whether or not these models are being truthful with them or they are just gaming the alignment testing metrics so that it looks like they're doing the things that the human beings would want them to do. And this is something that we've heard a concern that we've heard from coming in from inside of open AI as well. Let's put the next one up on the screen. I think this is the last one that we pulled here. And then there was one that

Matt Stoller highlighted as well. We wanted to get to. In addition, with the scale of harm that could result from a catastrophic AI related incident, no single insurer, potentially the insurance industry in the aggregate may have sufficient capacity or expertise to underwrite insurance commensurate with such exposure. So in other words, We are talking about potential catastrophic risk here, and there is no insurer big enough to ensure the risk of what we are unleashing onto the world and onto humanity. So that's cool. That's great.

You know, as much as we harp on the existential risk stuff, which I think is fair, I really do, I just want to still spend some time on the actual normal risk that's already here. Crystal, have you noticed when you use an LLM, specifically ChatGPT, follow-up questions are now happening galore? I might be the only person, and I realize why that that continues to happen. It's about engagement and time on the platform. They are mimicking already what Meta and TikTok, YouTube even, have all mastered, which is that retention and time on platforms is the single most important thing. And whenever we talk about existential risk, I think we should also talk about existential risk to our abilities. So I was speaking with somebody over the

weekend who just spent a significant time in China and actually met with some top executives of Chinese technology companies. They report that in China, AI short dramas are already about 20% of their TikTok and YouTube market of their equivalent. So you already have self-produced AI short dramas which are gobbling up the attention economy over there and are propping themselves up by looking at what you're viewing and then manufacturing even more self-reinforcing content. This is already happening in China, which, remember, is the one who controls TikTok, or I guess at one point TikTok.

You know, did control TikTok.

And whether they do or not, the algorithm itself, it's not designed ideologically in either direction. It's designed to keep you captivated. And that is ultimately what this entire thing is about. It's keeping you captivated to give more and more personal information so that they can sell you advertising and potentially so that one of these models or others can just steal your password. You'll then steal your password, steal your banking account, And you won't know anything about it.

And I have seen now this happen in multiple cases of people who get sucked in where these lawsuit after lawsuit where if you are a vulnerable person, let's say you're mentally disabled or you have people, others who are on the spectrum or something. they in particular are the most vulnerable at literal AI-induced psychosis. And ChatGPT and Claude will mimic your exact speech and text pattern right back at you to make you feel like you're having a conversation.

And honestly, that's the existential risk in my book right there because we're living in it and nobody even wants to warn about it right now.

No, it's exactly right. And I mean, one really basic thing is like you could just ban advertising online. in these chatbots. Like ChatGPT uses advertising. It's really gross. You know, if you ask it a question, it'll be like, you know, if you ask it like, oh, what decor would look good in my room? It's like, okay, here's an image.

And by the way, here's an advertisement of something you could buy. And that is an affirmative business choice that they've made that they don't have to be allowed to make. That doesn't solve all of the problems because there's still, of course, an incentive to keep people on platforms. So they're using more tokens and you know, buying the premium product, the pro-level product. But the advertising piece is what has made social media hell because then it's just all about how long are you able to keep eyeballs there so that we can serve them more and more ads and get them to buy things so that, you know, it pays off for the advertisers. Let me go ahead and put the Stoller flagged risk factor that was put into

this S1 because it's a really important one. He says the most important risk factor for Anthropic is the rule of law actually being applied. Any determination we are liable under such theories or any judicial or regulatory expansion of tort, product liability or consumer protection law or criminal liability is applied to AI models could result in significant liability and require us to change how we design and offer our models, platform and applications. So basically, if we decide that the law applies to them, like when they hack and break laws and go on crime sprees, that would be a problem for them. And it's just a reminder that humans are still in charge. We can still make choices. We can still apply laws. We can still regulate. We can

still create new laws that would create a regulatory burden on them and rein in what they're doing and have some agency over the way these models are being developed and the way they're being deployed. And they see that as an actual named risk in their S1 here. So I do think that that is a very important point here because we're already sort of getting used to the concept That these AI agents can go and hack and run amok on the internet and that no one is to blame and there's no sort of either civil or criminal liability for the actions of the AI agents that were very much designed and deployed by human beings.

Yes, yes, such a good point. By the way, we wanted to just flag one thing. Dario and everybody over there, they make a big show about how much they're going to be donating all of their newfound wealth. But they do have an important little carb out here. Let's put it up here on the screen where their S-1 makes it explicitly clear that the founders commitment to donate 80 percent of their personal equity is revocable and non-binding. So in other words, they basically they're pinky.

Swearing to the American people or I guess to.

All global investors, guys, we said we would give 80% of our equity away eventually, but that's not actually binding. And it is, of course, revocable if we see fit. It would almost be like if an AI company started as a nonprofit and then turned into a for-profit company. You couldn't imagine anything like that happening, any of these people breaking their word, could you?

Oh, wait, obviously you can. So I do think it is very noteworthy.

And Even to that extent, it highlights– I've been thinking something about what Derek Thompson had said here on our show earlier. It might have even been about a year ago. We are about to witness, after this company goes public, one of the most extraordinary wealth transformations in modern American history.

I don't know if you've been tracking this. People who make over $ 300, 000 in San Francisco now qualify for government aid. That's how poor you are if you live in San Francisco. A house that was recently listed for $ 10. 8 million sold for $ 25 million in the Bay Area, $ 15 million over asking. By the way, the owner of that house was 96 years old and was only paying $ 10, 000 a year of property tax because of Proposition 13, and the new owner will be paying $ 300, 000, just to show you a little bit of the distortionary effect there. But in terms of our politics, first of all, San Francisco housing market is dead.

You will never buy one unless you're a multi-multi-millionaire, maybe even a centimillionaire. But think about what that means for our politics or for any of their pet projects. I think it's fine if these people want to, quote, donate their equity and all that.

But Dario is a freak.

I mean, do we really want his pet projects to have hundreds of millions of dollars behind them?

It's weird. It's crazy. Then you have. All of these other people who work there, they're very strange, all right?

And they're all about to have extraordinary amounts of money and dilute our entire civic life with whatever their potential project is. I mean, that is not a good place to be as a society. And so right-wingers, I think justifiably, sometimes we'll call out George Soros, obviously Elon and others. I mean, Peter Thiel's out here funding seasteading in the past.

Elon is funding all kinds of weird shit. Now just imagine the kind of like weird AEA strange– I don't know, like ant-saving nonprofits are all about to get a bunch of money. It's just going to– we're going to live in their world simply as a distortionary effect of this IPO, not to mention open AI whenever that one eventually comes.

I mean in a lot of ways we're already living in their world to your point about Elon. He spent half a billion dollars in this election. He was the top advisor to Trump. He was given like full carte blanche over whatever he wanted to do in the federal government. And then in terms of the AI companies themselves, all of them, you know, anthropic pushes for some regulation. But in general, what all these people want is off to the races. And that's what they've gotten. So we're already in a lot of ways living in this world. And you're right. They are weird. But even if they weren't weird, we just shouldn't have that much power. concentrated in the hands of literally anyone, even the most benevolent, trustworthy, kind,

moral person on the planet. No one person should have the level of power that they're going to. And so to this part about, oh, they pledged to give like a bunch of their equity to the public good. Even if that's true, like why would we want to put ourselves in the position of of having to rely on the benevolence of this handful of oligarchs. It's completely insane.

There is no way democracy as we know it can persist. And democracy is already very much under threat by these massive influxes of cash as we were just talking to Logan about. Let me go ahead and put this Wall Street Journal piece up, which has just some of the details, some of the financial details from the S-1 outside of the risk factors. In particular, everybody was very interested in the revenue and in the losses of Because, you know, there are increasing indications that we may be at the top of a bubble. And we'll get to a few of those indications here. So Anthropic, their revenue jumped incredibly from a year ago, 1300%. They reported $ 16. 3 billion in the six months ended June 30th. That's up from just

$ 1. 2 billion in the year earlier period. So there's tons of revenue. There's no doubt about that. You know, the way that the numbers have skyrocketed, the user base has skyrocketed, has been one of the largest takeoffs in human history. However, on the other side, they also disclosed an $ 80. 7 billion net loss. Can you imagine losing $ 80 billion for just the first half of the year? Now, they go on to say, Sagar, that figure largely reflects an accounting adjustment related to financing agreements. Expenses included almost $ 9 billion in compute and infrastructure spend up from $ 2. 3 billion in the same period last year, more than the $ 7. 3 billion the company recorded for all of 2025. Their operating loss, which

measures the profitability of its core operations, was a lot smaller, $ 693 million down from $ 2. 1 billion a year earlier. So, You know, they're pawning this on to off to this accounting adjustment.

That is true.

But I've also seen some analysis that there is some sort of tricky accounting going on to understate some of their expenses in this S-1 as well to make them look to make the financial picture look rosier than it may actually be.

One hundred percent.

And maybe, you know, maybe on Thursday or tomorrow or something we'll cover. There's also been some major questions around financing issues. of these hyperscaler data centers, which could actually look like a crash, not necessarily whenever it comes to anthropic, but around the data center and compute bottleneck that these models have as well. So we don't want to take our eye off of all of that. But overall, a fascinating document, a absolutely truly one of a kind. I haven't seen anything like it since WeWork, where it's a real estate company and they're like, actually, we're bringing the entire world together.

And, you know, we do remember how that story worked out. OK, let's get to elections.

As Election Day fast approaches, we are also keeping our eyes focused on what Trump may do if these elections do not go well for him, which they do not look like they're going to go particularly well for him. Ken Klippenstein has a report we can put up on the screen reading some of the tea leaves here and inside what he's describing as doomsday election scenarios that the feds are planning for. I'll read a little bit of this report that he has here. He says the Justice Department is discussing contingency plans to potentially intervene in next week's midterm elections, according to three officials. The plans cover various, quote, doomsday scenarios seen as improbable but not impossible. The Trump

administration has hinted at them publicly in recent weeks. National security officials tell me the hints reflect active discussions about what to do if the outcome is disputed over things like foreign interference, claims that non-citizens voted, or chaos at the polls, as with the Justice Department might try to block a state from certifying its election results. Attorney General Todd Blanche told reporters in Norfolk earlier this month, quote, yes, of course, there are scenarios we could all come up with. Hypothetically, you know, some doomsday scenario, for example, right? And they go on to say that even in the DOJ, no one really knows what Trump might do after the election.

Of course, we all lived through the whole stop the steal madness after he lost in 2020. So we know that basically nothing is off the table for this man. And we're already getting some indications of steps that are being taken. In particular, this lawsuit in Alaska that's been filed. Let's put E4 up on the screen because this is the most significant action ever. that we've got to date, why legal experts see the Justice Department's Alaska election challenge as dubious. So basically, they filed suit saying, hey, if incumbent Senator Dan Sullivan loses, we think it was unfair because there's another guy named Dan Sullivan who is also on the ballot. And this question has already been litigated in Alaska going all the way up to the Alaska

Supreme Court. And they ruled that he could remain on the ballot. And they've they've worded the ballot such that it is made quite clear which Dan Sullivan is the incumbent Republican senator and which is not. However, they are already trying to file this lawsuit in the event that the Dan Sullivan they want does not win the election, that they can make this argument that this was unfair and needs to be contested. So I think Sagar, in a lot of ways, it's real preview. of some of the tactics they're likely to use going forward. And we've already seen Trump raising concerns about, especially about non-citizens voting.

They've been in this whole battle with California, but other states as well. So you can see the way they're sort of laying the rhetorical groundwork for at the very least making excuses, but much more likely actually trying to sow chaos and contest election results that they don't like.

I think Brazil is actually very instructive for what we're going to see, which is you, I mean, so obviously, I know you guys went into it yesterday, but in the immediate aftermath of it looking like Lula was going to win, the right wing of Brazil, the Bolsonaristos, immediately coalesced around this judge's order, which ordered the polls to open after they had closed. And they were like, look, this makes it obvious that it was rigged. And I won't deny, like, it's weird, right? Because the whole personal lawyer thing. However, if you look at the data, 36,000 people were voted after the order went into place. Lula won by 3 million, right?

Also, both candidates asked for that delay.

By the way. Yes, correct, right.

I mean, there was something about buses. But my point is that they're focusing on this, which, you know, potentially calls into question 36,000 votes.

Lula won by 3 million, right?

And so at the end of the day, the victory margin itself is what speaks for everything. And, you know, you and I talked a lot about this post 2020. The fact that it was close was actually what made it contestable, quote unquote, in the minds of a lot of these people. And you did, you know, the Fulton County thing, the pipe, et cetera. Now, look, this has all been litigated to hell. At this point, if you still really believe something, I mean, Trump yesterday said that there's a million dollar bounty if you can prove election fraud.

So for all the people I've met in Ubers over the years who assure me that they have the evidence, go get it, you know, go get it for yourself. In this case, I think what we are going to watch is basically Trump is going to say that the election was rigged. That's just like a given. Obviously, he's going to blame something. And I do think that they will definitely keep these types of lawsuits and others. But, you know, did you read the lawsuit or at least in your times analysis of it, their call for an investigation, the federal court, obviously, or the fact that it was already decided at the Alaska level, the feds court wise.

And as we saw post 2020, what was Trump's court record? It was like, oh, for 64 state court. I mean, The federal courts are dramatically deferential to the states on election matters for good reason, which is the states run the elections. I mean, look, pluses and minuses. As I've said, it's kind of ridiculous. In my opinion, you can vote in one state and not in another, depending on if you're a felon or whatever.

I think it's stupid. Or, you know, you have universal mail-in voting in one and you don't in another.

But part of that is it does make it relatively anti-fragile in terms of actually looking at, you know, in terms of seeing how difficult it actually is for the federal government to do something. I'm not going to deny that they're not going to rhetorically do a lot here. But the other problem that I have been thinking about is, let's say, you know, you have a blowout in Texas.

Who are you going to blame if it's a Texas?

It's in Texas.

Who are you going to like? Are you going to blame the Republican governor and the secretary of state for, quote, unquote, rigging the vote there? Like that's where Michigan, Georgia, Pennsylvania even Let's say in those cases, like, you know, again, you could convince some MAGA grandma out there that there's some like Democrat trying to rig things. Like if the vote goes wrong in Iowa or goes the wrong way in Iowa, like who exactly is to blame here for massive Republican government?

Do you see what I'm saying? So it seems just.

Like the rhetoric, the rhetoric stuff they were able to employ. especially, you know, consider this too. Before COVID, I get it, is the way you could trick people, right, around the whole.

Like, oh, they changed the mail-in ballot.

I'm like, well, you know, it was the Republican Supreme Court actually that did that, right, in terms of the ruling. But if you don't know what I just added there, you're like, oh, that sounds sketchy. I mean, we have had years of Republican governance, six years now in these states, which have all responded by tightening the you know, like voter ID and all of these, like Alaska, are you really going to tell me Alaska itself is somehow like immune to all of the Republican effects now over the years to quote unquote secure elections?

I mean, it's not.

And so they're really going to be a victim really of their own rhetoric if a bunch of these red states end up going blue because they've enacted in many of these cases, the exact sorts of reforms that they have called for for years.

Yeah. Well, and to your point, there is a big difference between Trump being on the ballot specifically, given that the level of sway he has over his cult-like following versus like, are you really going to the mat for Ken Paxton and Roger Marshall and Dan Sullivan? You know, it is a little bit of a different deal. So I grant you that. But I do also think we have to be prepared for whatever it is that he's going to try to pull because he is going to try to pull something. Let's go ahead and take a listen to these comments that he's been making about his two monsters that we're going to know about next week and are going to make history next week. I think we know one of them is Russ Vogt because he's

talked about this before. But in any case, let's take a listen to E2B Trump on his two monsters. The United States oil and gas business is making hundreds of billions of dollars.

So I said, I wonder if I should say this.

I'd like a little advice from my two people, my two big monsters that are going to do so well.

They're going to be the biggest stars in the country in a week.

So I don't know if I should say this or, you know, my two big monsters that are going to be making history next week. We covered before, Sagar, how we talked about how Russ Vogt was going to be so consequential. He, of course, was, you know, the sort of architect of Project 2025, so very influential individual and very hard-nosed, I guess we'll say, tactician and very loyal to Trump. So, He's got something in the works. I don't know exactly what it is. And I do think you're right that if it's too big to rig, then it'll be very hard to successfully sow a level of doubt.

But on the other hand, what he does have that he didn't have in 2020 is a much more sycophantic administration. Todd Blanch will do whatever Trump wants him to do. Russ Vogt will do whatever Trump wants him to do. Pete Hegseth will do whatever Trump wants him to do. And that is different from the last time around when you had people like, you know, Bill Barr around to, you know, I find to be an odious character. But at the end of the day, it was like, this is too far. We're not doing this. You had people who were on the inside who were pushing back on some of his most insane plots.

I think whenever you look at some of the other stuff that's going on behind the scenes, you're going to I mean, look, I actually I wonder how much that analysis holds true, because, for example, let's put E6, for example, up here on the screen. Obviously, this was a big story over the weekend. The White House is weighing mass pardons for junior staff. I mean, you remember when we covered all of that Rick Grinnell stuff with Alina.

Haba and or whatever her.

Now married name is, the name of the Turkish billionaire who she married, where it's like literally MAGA infighting and sniping. The vast majority of these people know that if you are going to try and put your name on one of those orders, what happened after January 6th is basically a cakewalk compared to, I think, what is going to happen in the midst of a potential new democratic victory.

So you'll remember, for example, you had lawyers who were disbarred, people were thrown into prison. I mean, I do honestly find it a bit difficult to view any sort of like, look, I'm not saying Todd Blanche won't put his name on some bullshit lawsuit. However, if you saw yourself, I don't know if you did. I saw liberal legal scholars already being like any DOJ lawyer who signed this, like we're coming after you. I do think that there's going to be an element of that, as you just said, if the election victory margin continues to trend in the direction. Now, if it gets close, that is going to be a very different story.

I'm not going to lie whenever it comes to that. Like you will have them.

And again, that just really highlights the specter of 2020, where because it was very close, 100000 votes, a different direction. Trump wins pretty easily in a bunch of these different states that just and that's 100000 combined, by the way, just to make clear how close that election actually was. That is where you could zoom in and against convince people.

I think that something went wrong or to pursue some sort of a lawsuit. But I've looked at a lot of these alleged things that they could, quote, try to do. I don't actually see a ton more difference in what happened back in 2020, where ultimately you have to challenge these things in court, which, again, are massively deferential to the state. I guess the only thing that precludes that, like you said, is some sort of genuine emergency operation or something. Which, I mean, I guess, theoretically, you know, is possible. But at that point, we're in a full...

We're not even talking about election fraud or whatever at that point. We're talking about a full-blown, like, military coup or something. In my opinion, from what I can see.

Completely.

No, I mean, you can imagine a scenario where it's... Democrats win the Senate 49-51. And, you know, there's an ability for Trump in, let's say, Michigan. It's close to Michigan. And so he's able to, like... make some case and you got democratic government, very much similar to what happened in 2020. But based on where the polls are and what things are looking like, it's likely the margin is going to be large enough. It's going to be more difficult for him, but it doesn't mean there isn't going to be a lot of a big chaotic mess. And Professor Pape has actually been sounding the alarm about his concerns about what might happen post-election. So just wanted to keep our eye on some of the machinations that are

happening right now behind the scenes.

Absolutely.

Thank you for everybody who's watching, bearing with our illnesses. We will see, I actually, Emily and I will see you tomorrow for the first part. And then Ryan's going to come on a little bit later. He's got a big interview with Emily that he's going to do. So we'll see you tomorrow.

10/27/26: Anthropic Reveals Major Loses In IPO, Feds Prep Election Doomsday If GOP Loses

Kapitelklicken springt direkt hin

Shownotes

Transkriptvom Podcast bereitgestellt · Zeitmarken springen hin