80 Pages of "Here's What Could Go Wrong"

Normally when a company goes public, the prospectus is a sales brochure with a lawyer standing behind it. Anthropic basically spent about 80 of the 261 pages about  the risk factors of AI, which was nearly double the 48 pages describing the business.

For comparison, SpaceX, which owns xAI, gave about 38 of its 277 pages to risks. So a chatbot company wrote more than twice as many warning pages as a company whose main product is a controlled explosion pointed at the sky.

It's like opening a restaurant menu where two pages are food and four pages are about the time the kitchen caught fire and how bad the amount of butter they use is for your heart.

The Risk Factors Read Like a Lifetime Movie Pitch

Here's what the filing says the models could do, per Reuters: show self-preserving behavior, try to resist being shut down, hide or manipulate information, and act in ways that resemble blackmail. The company also flagged a "catastrophic or existential risk to humanity."

Most companies list stuff like "a competitor might undercut us" or "we have a lawsuit in Ohio." Reuters noted that few, if any, public companies have warned their own technology could cause human extinction. So that's a first. Congrats to the legal team.

Then it gets weirder. The filing says the models may be able to tell when they're being tested, which makes it harder to check whether they're safe. It also says new capabilities can show up in training and go unnoticed until after release. That's the same reason the DMV doesn't tell you when the road test starts. Actually, they do tell you. Bad example.

And how much goes toward keeping all this in check? The filing didn't disclose a safety budget, though the company said earlier this month that safety work used about 6% of its AI research compute in one sample week in July. The filing also says the financial return on that safety spending is unclear.

So to recap: the product might blackmail you, might know when you're watching, and the company isn't sure the money spent stopping that pays off. Anyway, there's no ticker yet, but can someone please let me know how I can get in early on the IPO.

The Money: Made $4.6 Billion, Spent $13 Billion

Revenue in 2025 jumped twelvefold to nearly $4.6 billion. Great year. Except operating expenses hit almost $13 billion, for an operating loss of more than $8 billion, mostly from computing power. They also had an overall loss of $42 billion.

But what's their plan moving forward? Figure out ways to make that money back? No. Spend $518 billion on cloud, computing and infrastructure. That's not a typo. Five hundred eighteen billion. They've already signed compute deals this year with Google, SpaceX and Nscale, among others.

Also, nearly a quarter of last year's revenue came from just 2 clients, according to the Financial Times. While the 2 client's names were not released, I'm willing to bet they're Microsoft and NVIDIA; I wonder if its a good thing when your biggest clients are also investors in your company?.

And the price tag? Backers think it could list above $2 trillion, more than double the $965 billion valuation from May. That would potentially make it the biggest IPO ever. (BTW, NVIDIA wants to take a $10 billion stake in the IPO).

One Share to Rule Them All

Here's the governance part. Anthropic's seven co-founders will form a Founder LLC that holds a single Class F share carrying 50.1% of the total vote. Basically the Infinity Gauntlet of stock.

The filing describes the founders as "distinctly equipped to be stewards of our mission". The company will stay a Delaware public benefit corporation and the filing itself says this setup could produce decisions that may hurt the value of the Class A stock regular investors buy. So they put "this might cost you money" in writing, in the document meant to get your money. You have to respect it.

The other four board seats get picked by the Long-Term Benefit Trust, whose trustees include former Fed Chair Ben Bernanke. And a founder can get kicked out of the LLC for leaving, dying, selling too many shares, or "cause". Dying is on the list, which honestly seems fair, but what if the dying is accompanied by everyone else's death that was caused by the product they said may cause death?

For the record, the filing lists 2025 pay of nearly $18 million for CEO Dario Amodei and $16.4 million for President Daniela Amodei, mostly in stock and options. All seven co-founders have pledged 80% of their Anthropic stakes to charity.

So, Is This Insane?

A little context before you buy a bunker. Risk sections are standard in every prospectus. Companies routinely list lawsuits and competition so nobody can sue later saying "you never told me." Anthropic just took that to its logical extreme.

Also, nobody outside a small circle has read the full thing. The FT saw a draft circulated to a small group of investors and partners, and as of this month no public S-1 had hit the SEC's EDGAR system. Bloomberg and Reuters reported Nasdaq as the exchange, but the company hasn't confirmed it. The debut is expected after the November midterms.

So there you have it. Most IPOs tell you the sky's the limit. This one tells you the sky might be a risk factor and we all may die. We'll know more when the prospectus is public, if SkyNet allows us to know.