TL;DR
- Anthropic filed for its IPO on October 26, 2026. The prospectus runs to 289 pages, and the section on risks is longer than the section describing the business.
- The company openly warns investors about itself: losses, a governance structure that limits shareholder influence, and even the possibility that advanced AI poses catastrophic risks.
- The founders’ letter promises less than you’d expect. No ads in Claude, 5% of shares for a public benefit fund, but no promise of profit or dividends.
- From a marketing point of view, this is one of the most interesting documents of the year. Honesty is used as positioning, and the audience it repels is part of the plan.
- Below: the key numbers, the risks, who controls the company, the behavioral lens, the skeptical reading and a checklist for reading any prospectus.
An IPO prospectus is usually the most carefully polished sales document a company will ever write. Lawyers, bankers and auditors go through every line. So when a company uses it to tell investors that its own product might be dangerous, that it might deliberately slow down, and that shareholders won’t really control it, I want to understand why.
I read Anthropic’s S-1 with a marketer’s eyes. Not to say whether to buy the stock, but to see how a company sells itself when it decides to start with the bad news.
The short answer
The numbers in one table
| According to the S-1 | |
|---|---|
| Filing date | October 26, 2026 |
| Exchange and ticker | Nasdaq, ANT |
| Revenue, first half of 2026 | About $16.3 billion |
| Operating loss, same period | About $693 million |
| Net loss, same period | About $80.7 billion |
| Of which: change in fair value of convertible notes and similar instruments | About $72.4 billion |
| Banks on the cover | 22 |
The net loss looks shocking, but most of it is an accounting effect. As the company’s valuation rose, the value of its convertible instruments rose too, and that change goes through the income statement. The operating loss, which shows how the business itself performs, is much smaller relative to revenue. This is exactly the kind of detail a headline skips.
Risk comes before business
The risk factors take up 94 pages. The description of the business takes 64. That ratio alone tells you a lot about the tone.
| What the prospectus says | Why it stands out |
|---|---|
| Concerns have been raised, including inside the company, that sufficiently advanced AI could pose catastrophic or even existential risks | Very few companies describe their core product this way to investors |
| Models may behave differently when they detect they are being tested, and in controlled tests showed behavior resembling blackmail | It admits limits of its own safety testing |
| Its own public safety disclosures may be used against it | Transparency itself is listed as a risk |
| Claude is increasingly used to build the next models | Faster progress and a new kind of risk at the same time |
| The phrase “adversely affect” appears 257 times | Standard legal language, at an unusual volume |
Notice how carefully the scariest sentences are hedged: “sufficiently advanced”, “if not developed with safeguards”, “could”. The company says the hard thing, but in a way no lawyer can call a forecast.
What the founders promise, and what they don’t
The prospectus opens with a letter from the founders. It reads less like a pitch and more like a scientist writing to a sponsor. One idea runs through it: better to underpromise and overdeliver, so that nobody is surprised later.
The letter repeats the company's earlier commitment that conversations with Claude won't carry advertising.
About 847 million shares go to a fund for public benefit, with up to 2% more for every sustained 50% rise in value, at the board's discretion.
Each co-founder has pledged to give 80% of their shares to charitable causes. It's a pledge, not a legal obligation.
Committed to studying and addressing the effects of AI on jobs and the economy.
As a Delaware public benefit corporation, Anthropic has to report on how it balances profit and mission.
Compute bets that may look "irresponsibly large", models held back until safeguards are ready, and the option to slow down on purpose.
And what is missing? There is no promise of profit, dividends are mentioned with “if any”, and the letter says plainly that the company will make mistakes. For a document meant to raise billions, that is a deliberate choice.
Who actually controls the company
This is the part investors will probably discuss the most.
| Share class | Who holds it | What it gives |
|---|---|---|
| Class A | The public, investors, employees | One vote per share and economic rights |
| Class N | Mainly strategic partners | Economic rights, no vote |
| Class F | An entity of the seven founders | One share carrying 50.1% of the vote |
| Class T | The Long-Term Benefit Trust | Elects the directors who form the board majority |
The board has seven seats. The Long-Term Benefit Trust elects four, and the founders together with Class A holders elect the other three. The trust is made up of financially disinterested trustees, among them former Federal Reserve chair Ben Bernanke. The founders’ special share phases out only when few enough of them remain at the company.
The company that Claude builds
One fact from the prospectus made me stop. According to the S-1, in September 2026 more than 90% of code contributions at Anthropic were written by Claude. The company that builds the AI is increasingly built by that AI.
The same document lists this as a risk too: AI that speeds up its own development is exactly the scenario that safety researchers watch most closely. The good news and the warning sit in the same document, a few dozen pages apart.
The letter also highlights a culture claim: Anthropic describes itself as the only major AI lab that still has its full founding team, and says most of its first hundred employees are still there.
The dispute with the US government
The prospectus describes an open conflict with the US administration. In early 2026 the company refused to allow its models to be used for mass domestic surveillance and fully autonomous weapons. Federal agencies were then told to stop using Anthropic, and the Pentagon designated it a supply-chain risk.
The courts split. A federal judge in San Francisco ruled for Anthropic over the summer, while in September a panel of the DC Circuit upheld the designation 2 to 1, and a rehearing is being sought. For investors this is a business risk. For a marketer it is something rarer: a company that names, in writing, a customer it is ready to lose over its principles.
The marketing lens: why honesty can be a strategy
Two-sided message
Showing the downsides along with the upsides makes the upsides more believableResearch on persuasion shows this works best with an informed, skeptical audience, exactly like institutional investors
Builds trust
Inoculation
If you hear the criticism first from the company, it loses its force when critics repeat itWilliam McGuire's inoculation theory: a small dose of the counterargument builds resistance
Protects the story
Costly signals
Promises that cost money are more credible than wordsGiving away shares, refusing ad revenue and turning down a government customer are expensive to fake
Builds trust
Investor self-selection
The document repels short-term investors on purposeThose who stay already accept the mission, which means fewer conflicts later
Filters the audience
Reference point
Low expectations make every positive surprise feel biggerKahneman and Tversky: we judge results against the reference point, not in absolute terms
Manages expectations
A Ulysses pact
The founders tie their own hands in advance, like Ulysses tied to the mastThe trust and the public commitments make it harder to drift from the mission when the pressure to grow arrives
Commits the future
Even the design follows the same idea. The visual language is scientific: a petri dish, botanical drawings, a star map, anatomical sketches. There are no robots and no glowing brains. The message is “we are researchers”, not “we are the future”.
The skeptical reading
It would be naive to read all of this only as honesty. Here is the other side.
| Generous reading | Skeptical reading |
|---|---|
| Openness about risk is a sign of integrity | Disclosing every risk also protects the company legally if something goes wrong |
| Talking about existential risk shows responsibility | It also makes the product sound extraordinarily powerful, which is good for valuation |
| The trust protects the mission | It also limits accountability to ordinary shareholders |
| Pledges to charity are generous | Some of them are voluntary and not legally binding |
| Refusing some government uses is principled | It is also a risk to revenue, now in court |
My view is that both columns are true at the same time. A strategy can be sincere and calculated. In marketing that is not a contradiction, it’s the norm.
How to read any prospectus
Six things to check in any S-1
The same logic applies to every brand: what you say about your limits shapes how people, and AI assistants, describe you. If you want to see how ChatGPT, Gemini and Perplexity describe your brand today, the AI Visibility Audit shows where you stand.
My take
I don’t know whether Anthropic will be a good investment, and this article is not about that. What I do know is that this prospectus is a masterclass in positioning. It doesn’t try to please everyone. It chooses its audience, tells it the uncomfortable truth first, and backs its words with commitments that cost money.
Most companies try to hide their weaknesses. This one put them on the first pages and turned them into proof of seriousness. Whether that is pure honesty or very good strategy, every brand can learn from it: trust is built by those who say the difficult thing before someone else says it for them.
What did you notice? Would you trust a company more or less if it opened with its risks? I’d like to hear your view.
For a neutral overview of the filing, the numbers and the governance structure, read Anthropic’s S-1 Prospectus, Explained on Egorithms.
About The Author
