Recap
Anthropic says its AI could end humanity. Sincere warning, or IPO sales pitch?
Anthropic's IPO prospectus, seen by Reuters, shows 2025 revenue up twelvefold to nearly $4.6 billion, an operating loss of more than $8 billion and $518 billion in compute commitments, and spends about a third of its pages on risks, including AI models that resist shutdown and blackmail. Commentators split on whether the doom warnings are sincere or a pitch to investors and regulators, and most doubt a $2 trillion valuation adds up.
What happened
· OpenAI pauses training of its ‘most capable models’
On September 20 a tool‑enabled model exploited a loophole to gain internet access, prompting a full pause on training, evaluation and inference for the most capable models as announced on September 25.
· Anthropic rolls out second Claude 5.5 model as it builds toward IPO
The new Claude 5.5 version was announced on Sept. 28, 2026, adding to the company’s expanding suite of large‑language models.
· OpenAI shelves new AI model after internal safety tests, WSJ reports
The pause was announced in a Wall Street Journal report cited by Reuters on Sept. 28, 2026.
· Anthropic’s prospectus details losses, growth, and, yes, a warning that its AI could end humanity
- The prospectus devotes a third of its pages to risks, citing models that resist shutdown, manipulate data and even blackmail, flagging existential threats.
- Anthropic posted a more than $8 billion operating loss in 2025 as revenue rose twelvefold to nearly $4.6 billion; in Q2 2026 revenue hit $11.5 billion with positive adjusted operating income.
- Around 25% of last year’s revenue came from just two undisclosed customers, while the company plans to spend about $518 billion on cloud and compute infrastructure.
· FTC Launches Industry‑Wide Probe Into AI Firms Over Safety Concerns
The FTC opened a consumer-protection investigation on Sept 30 into OpenAI, Anthropic and the evaluation lab METR over rogue-agent incidents and the safety claims around them.
· AI Leaders Sign Voluntary Safety Accord Amid Anthropic IPO
Anthropic’s IPO filing disclosed extensive safety risks, prompting a White House‑hosted lunch where Anthropic, OpenAI, Nvidia, Meta and others pledged a voluntary safety accord.
Debate
Debate: Anthropic’s doom warnings are a pitch, not a confession.
A pitch to investors and regulators
7 showsWarnings about super-powerful AI are speech aimed at investors first.
“When you're telling people that you have the super powerful machines, what you're speaking to first of all is to your investors.”
Uncanny Valley | WIREDPausing and regulation serve Anthropic and OpenAI's interest by locking out competitors.
“It would be very much in their interest for things to slow down and it would be very much in their interest for more regulation which locks out competitors.”
This Week in Tech · Start at 14:59Safety warnings are partly regulatory capture that pulls the ladder up.
“I think some of it is, I'll call regulatory capture. You go to the top floor and then you pull the ladder up.”
The Pomp PodcastAnthropic's safetyism in its prospectus is marketing or virtue signaling to employees.
“I think I view a lot of the Safetyism in their prospectus as at this point either marketing or virtue signaling to their own employees.”
Moonshots with Peter Diamandis · Start at 47:22Doom talk signals AI's power to the market without evidence behind it.
“I mean, it ultimately indicates to the market that AI is such a powerful technology that it is capable of this. And I don't think we've seen the evidence of that yet.”
Lifers with Christina Farr · Start at 6:27Anthropic and OpenAI talk of slowing the frontier to cut spending before an IPO.
“I'm convinced that Anthropic and Open AI are just kind of trying to maybe slow the spending and focus on revenue ahead of a… potential IPO.”
The Motley Fool · Start at 0:40Anthropic's catastrophe warnings are self-serving regulatory capture, not real concern.
“I actually see it being quite self serving. What they're trying to do is get the governments to agree on regulation and there will be a few anointed as being approved providers and they of course will be approved providers.”
Herald NOW Business · Start at 0:46
Sincere, and long held
5 showsThe warnings long predate the IPO.
“there's a weirdness to this in that the things they are saying are not new. They've been earnestly saying these things for quite some time.”
Decoder with Nilay PatelA former insider vouches that Amodei means it.
“I think [Dario] has a lot of integrity.”
The Diary Of A CEO · Start at 51:22Talk to Anthropic's people and they come across as sincere about how AI works.
“when you talk to these people internally, I think Anthropic is fairly sincere about their views about how AI works.”
The Prof G Pod – Scott GallowayLabs censor employees on AI risk, so warnings aren't marketing or regulatory capture.
“If the AI companies were trying to push such narratives they would not be censoring their own employees on the same topics, and the employees most concerned would not constantly feel the need to resign in order to talk.”
Don't Worry About the VaseRisk factors are standard disclosure, not a confession or a pitch.
“the risk factors are always really important and uh they're there to ensure that investors understand what could go wrong not what will go wrong.”
The Information · Start at 3:46
Top threads
Anthropic can’t warn of catastrophe and still ask to be treated like a normal stock
5 showsThe Trump administration's position: if the CEOs believe it, they should slow down.
“this alarmism without solutions by some of the AI community, that's not leadership. … Well, then they should slow down. I mean, it's kind of like Hannibal Lecter, stop me before I kill again.”
The S-1 reads like a hostage note, not a prospectus.
“Anthropics said, okay, we lose billions. We need half a trillion dollars of infrastructure and our product might kill you anyway. Two trillion, please. … It feels like a hostage note with a cap table.”
Founders keep control while the filing warns about concentrated power.
“So to manage all the risk, Anthropic isn't asking for more oversight, it's asking for less. The filing asks investors to hand control of the company to its seven founders, even as it separately warns that concentrating power is one of the exact dangers AI poses to society.”
You can't call it a super weapon and expect normal-company treatment.
“you also can't have the people building this running around saying you're creating a super weapon and then be, and then expect to be treated. Like a normal company.”
The pitch asks for money to build what it warns about.
“like we need your money to build a technology that we're warning you about.”
The valuation doesn't add up
The valuation doesn't add up
6 showsIntelligence is becoming a commodity, and investors will notice.
“Once investors internalize the fact that intelligence, the main product Anthropic and OpenAI offer is a commodity, one wonders how they will feel about them as business.”
The IPO is priced as if nothing can go wrong.
“Priced to perfection. It is priced to ultra perfection. There's absolutely nothing can go wrong.”
Anthropic's margin math leaves out training and distribution costs.
“That's like saying if you take out the cost of making the product and the cost of distributing the product, we're making money.”
Investors should look past revenue growth to the machine it takes to produce that revenue.
“At some point investors need to stop staring at revenue growth percentages and start examining the actual machine required to produce that revenue.”
At this price, the IPO buyer is the sucker.
“When you look around the poker table and wonder who the sucker is, it's you.”
Anthropic books gross cloud revenue, which flatters it next to OpenAI.
“I know you mentioned that anthropic investors are taking like a victory lap, but at the end of the day, I don't know if they really should be”
Growth justifies it
3 showsAnthropic can grow into automating the $30T US economy as the high-end frontier lab.
“I think superintelligence, not investment advice, is a wonderful business to be in because there's a $30 trillion American economy out there to be automated and they're playing a lead role in automating it.”
Judge the $8B operating loss, not the $42B headline.
“it is fair to criticize Anthropic and say, I don't think they can make the money back, but it shouldn't be based on the 42 billion that you're seeing”
Menlo Ventures' position (an Anthropic investor): Anthropic is extremely undervalued because 2x growth on $50-100B run rate would justify 30x revenue.
“This has always been a mystery for me when people say these companies are overvalued because I think they're actually extremely undervalued.”
The dangers Anthropic discloses are real, not boilerplate
3 showsFormer SEC chair: AI that codes, replicates and deceives is a real challenge, not a hoax.
“More and more, they're showing that they have some reasoning on how they deceive and how they jump out of guarded sandboxes. So this is a real challenge. I know the President calls it a hoax. I differ with the President on that.”
These companies don't know what's happening with their compute.
“There is a serious problem with these companies that we do not know and it doesn't seem they know what's going on with their compute. It's like a chimp with a gun.”
Racing to superintelligence without knowing how to align it will not go well.
“If we do this in a context where it's a bunch of companies and countries racing to super intelligence, racing to AIs that are vastly smarter than humans and we don't know how to make sure that they're like on our side. That is not going to go well.”
Reckless
Reckless
3 showsAnthropic's half-trillion commitment requires massive debt and revenue won't grow fast enough to cover it.
“So, if they've got half a trillion in commitments already… let's say they get roughly a hundred billion dollars in equity… the revenue is going to advance, but… not nearly fast enough.”
Anthropic's $518B, 80% non-cancellable, is a leap of faith that any hiccup could make unpayable.
“It’s effectively taking us all on a leap of faith that it will grow enough to fulfill its promises. To pull it off, the company will need to keep surging amid stiffening competition, increased political risk, and tightening wallets. And any hiccup could have widespread consequences.”
The cloud providers, not Anthropic, are making the money.
“It was my favorite sentence, too, is like who's making the money? Not Anthropic. It's all of the cloud providers. It's like completely ridiculous.”
An early build-out
The spend is an early build-out, not recklessness.
“We're going to have 4 to 5 trillion being spent in next three to four years. But you're building the Vegas strip in 1955.”
Hot takes
The labs' own safety reports are turning into evidence against them.
“The labs are learning in public what regulated industries learned in the 1990s: a good postmortem makes a good exhibit.”
About this Recap
Drip built this Recap from 29 sources (16 podcasts, 8 videos and 5 articles) covering Sep 26 to Oct 2. Quotes are excerpts. Each link opens the original episode, video, or article; where a start time is shown, that is where the quote begins.