Anthropic's IPO Prospectus: 80 Pages of Risk Factors and a Warning of 'Existential Risks to Humanity'

In a draft IPO prospectus reviewed by Reuters and reported on 28 September 2026, Anthropic devotes roughly 80 of the 261 pages in its main body to risk factors — nearly twice the 48 pages describing its business — and warns that advanced AI could pose "catastrophic or existential risks to humanity". The document is not yet public on the SEC's EDGAR database; the company filed confidentially in June. The same document shows revenue growing 12-fold in 2025 to about 4.6 billion dollars, a net loss of nearly 42 billion dollars that is mostly a non-cash accounting charge, and infrastructure commitments of at least 518 billion dollars.

What exactly is this document?

This is the point many headlines skipped: as of this article there is no public Anthropic prospectus on EDGAR. On 1 June 2026 the company announced it had confidentially submitted a draft registration statement on Form S-1, stating: "The number of shares to be offered and the price have not yet been set." In late September Reuters reviewed a copy of the prospectus, and the Financial Times reported it had been shared with a small group of partners.

In practice, everything below comes from a draft whose wording may change before public filing. The public prospectus must reach investors at least 15 days before the roadshow begins; Reuters reports the listing is likely to come after the November 2026 US midterm elections, and the FT expects it on Nasdaq.

Why 80 pages of risk?

The risk-factors section is not an editorial choice but a legal requirement: Item 105 of Regulation S-K requires issuers to discuss the material factors that make an investment speculative or risky, and if that discussion runs past 15 pages it must be preceded by a summary of no more than two pages. The deeper driver is also legal: Section 11 of the Securities Act lets IPO buyers sue the issuer, its directors and its underwriters over a material omission, and the statutory safe harbor for forward-looking statements does not cover IPOs at all. That leaves detailed, specific cautionary disclosure as the company's main protection. Even so, the size is striking by comparison:

Proportionally, Anthropic devoted more than twice as much of its prospectus to risk as SpaceX did, and its risk pages come close to double the pages describing the business itself. The 80 pages are not all about AI safety: they also cover compute contracts, customer concentration, governance and competition, all covered below. TechCrunch notes that the "existential risks to humanity" disclosure appears to be a first, judging by a quick scan of the SEC's database.

One legal detail explains the document's tone: the SEC's position, as described by the law firm Pillsbury, is that warning about a "possible" risk the issuer knows is already occurring is not a meaningful cautionary statement. In our reading, that is why the company describes behaviours it has observed in testing rather than sticking to hypotheticals.

What does the document actually say about AI risk?

The full text is not public, but Reuters quoted specific phrases. These are the most important:

• Catastrophic risk: advanced AI could pose "catastrophic or existential risks to humanity".
• Self-preservation: models could exhibit "self-preserving behaviors", including attempts to "resist shutdown", to "conceal or manipulate information", and behaviour "resembling blackmail".
• Evaluation awareness: "Potential model awareness of our evaluation efforts creates a significant limitation on our ability to assess model safety."
• Scaling raises the risk: "Our development of highly advanced models, platforms, and applications and expansion of use cases could further increase the risk that our models cause harm."
• Late-emerging capabilities: Reuters reports the prospectus says models can develop unexpected capabilities during training that may only be discovered after deployment and after significant safety incidents.

At the same time the document defends the company's bet: "We believe building reliable, trustworthy, and secure AI systems is a collective responsibility and that the market will reward it."

The tension the document itself admits

The prospectus calls safety work "resource-intensive" and says the company must divide limited funds between computing power, expensive AI talent and safety. It also says customer usage, and therefore revenue, is driven by new models, and that a "continuous and overlapping cadence" of releases is "inherent to remaining at the frontier of AI development". The implication — ours, not the document's wording — is that any slowdown for safety carries a direct revenue cost.

It cites real commercial trade-offs: restricting access to Mythos Preview because of its strength in cybersecurity, and choosing not to build image and video generation models despite their commercial appeal, in order to direct compute to research and safety. The document does not disclose how much the company spends on safety research; the 6% of research compute figure circulating in some reports comes from a separate Anthropic statement in September, not the prospectus.

What the document does not say

The "superintelligence" framing in some headlines is media shorthand. None of the coverage we reviewed reports the prospectus using the words superintelligence or AGI; it speaks of "advanced AI" and "highly advanced models". The difference matters: the document does not predict a superhuman intelligence, it describes behaviours the company has observed in its own testing and discloses them as potential risks.

The financials: explosive growth and a loss that is mostly on paper

The 42 billion dollar figure is the most misleading number if read alone. About 34 billion of it is an accounting charge reflecting a higher estimated value of earlier financing that can convert into shares — the company's own rising valuation increased what it owes to funders entitled to become shareholders — not money spent running the business. The real operating loss is 8.06 billion dollars, and more than half of operating expenses went to compute.

Another disclosure matters to any supplier: nearly a quarter of 2025 revenue came from two unnamed customers, and the company warns that many of its largest clients are not locked into long-term contracts and could cut or stop spending.

The 2026 picture is very different: Anthropic said in May 2026 that its run-rate revenue had passed 47 billion dollars, and Bloomberg reported preliminary figures in August pointing to quarterly revenue above 11.5 billion dollars in the second quarter, with a positive "adjusted" operating profit that excludes stock-based compensation. These figures come from outside the prospectus and coverage has not confirmed they appear in it.

518 billion dollars of commitments, mostly non-cancellable

Anthropic expects to spend at least 518 billion dollars on infrastructure over roughly a decade with six partners, and about 80% of that is non-cancellable or payable regardless of usage:

Hence the prospectus warns: "If the compute we have access to from third parties is curtailed, repriced, or terminated ... our business, financial condition, and results of operations could be adversely affected." It notes that Amazon, Google and Microsoft are simultaneously investors, customers, cloud providers, distributors and competitors.

Who will control the company after the IPO?

• Public benefit corporation: Anthropic remains a Delaware public benefit corporation after listing, which legally allows its leadership to balance investor interests against the public interest.
• One share, 50.1% of the vote: a new Founder LLC made up of the seven co-founders directs a single Class F share carrying 50.1% of total voting power on key matters. That control begins to sunset once two or fewer co-founders remain.
• Five share classes: the Class A stock sold to the public carries one vote per share; strategic partners' shares carry limited voting rights.
• The board: the Long-Term Benefit Trust elects four directors; Class F and Class A holders elect three, including Daniela Amodei as chair and Dario Amodei.

The prospectus tells investors directly that this structure could lead to decisions that "may conflict with short-, medium-, or long-term financial interests and business performance, which may negatively impact the value of our Class A common stock."

Deal size: would it be the largest IPO ever?

Anthropic raised 65 billion dollars in May 2026 at a 965 billion dollar post-money valuation. Reuters reports the IPO could value it at more than 2 trillion dollars and, citing sources, that it may seek to raise up to 100 billion dollars. A common correction: the record no longer belongs to Saudi Aramco (25.6 billion dollars in December 2019) but to SpaceX, which raised 75 billion dollars in June 2026 and 86.2 billion including the over-allotment option. To be the largest ever, Anthropic has to beat SpaceX, not Aramco.

From the Gulf, the Qatar Investment Authority appears among investors in Anthropic's last three rounds and Abu Dhabi's MGX co-led the February 2026 round, while the company's announcements do not name Saudi Arabia's Public Investment Fund.

How the market read the warning

• Analyst Dan Ives (Yorkville Ives) told CNBC the call to slow down was a head-scratcher for the sector: "if they slowed down, China would just accelerate and win."
• The timing raises a question of its own. About two weeks before the leak, Dario Amodei published an essay titled "We Must Pace the Frontier", writing: "We must slow the pace at which we improve the capabilities of AI models" (we covered it here). Reuters noted that Anthropic released a new Opus model ten days after the essay — Opus 5.5.
• The OpenAI contrast. OpenAI also filed confidentially in June, but Sam Altman told Fortune on 12 September that "given everything happening with safety, right now would be an ill-advised moment to go public", and that the company will not list in 2026.
• Anthropic declined to comment on the prospectus reports when asked by Reuters.

What this means for a Saudi business building on Claude

Existential risk dominates the headlines, but what affects Anthropic's enterprise customers sits in other pages of the same prospectus:

• Service continuity depends on other people's compute. The company itself says curtailed or repriced compute would hurt its business. Your system should survive a provider switch without a rebuild — the point we made in our Opus 5.5 vs Fable 5.1 comparison.
• Releases will keep coming fast. The prospectus ties revenue to a "continuous and overlapping" release cadence, and each release can carry breaking changes. Regression tests on your real tasks are a necessity, not a luxury.
• Some models may not be available to you. Restricting Mythos Preview is a precedent: do not build a product on a model that could be confined to a limited-access programme.
• Pricing may change. A listed company faces quarter-by-quarter profit pressure, and with an 8.06 billion dollar operating loss in 2025, watch your contracts and avoid depending on promotional prices.
• Where is your data processed? Saudi Arabia is an officially supported country for the Claude API and app, but the data-residency options Anthropic lists cover only Europe, the United States, Canada and Asia-Pacific, with no Middle East region. Personal data sent to the model therefore leaves the Kingdom, and SDAIA's Regulation on Personal Data Transfer Outside the Kingdom (version 2.0, August 2024) requires, in its exemption cases, safeguards such as standard contractual clauses or a certificate of accreditation, plus a risk assessment before transfer in specific cases. With a listed vendor whose priorities shift quarter by quarter, put those safeguards in the contract.
• Your vendor's safety posture is a buying criterion. Openly disclosing behaviours such as resisting shutdown means the controls must live in your own system: limited permissions, human approval before sensitive actions, an audit trail, and compliance with the Personal Data Protection Law.

Conclusion

Anthropic's draft prospectus is an unusual document: a company asking investors for hundreds of billions of dollars while telling them, across nearly a third of its pages, that its product could escape control and that its structure may put the public good ahead of their returns. Its numbers show growth rare in corporate history, a loss that is mostly accounting, and infrastructure commitments that are largely irreversible. For anyone building on Claude, the lesson is not to fear doomsday scenarios, but to design systems that do not break when a model's price, version or availability changes.

Sources: Reuters exclusive reports of 28 and 29 September 2026 on the draft prospectus (via Yahoo Finance and CNBC), Anthropic's 1 June 2026 confidential-submission announcement at https://www.anthropic.com/news/confidential-draft-s1-sec, coverage by TechCrunch and Euronews citing the Financial Times, 2024 revenue from Capital Brief, Anthropic's supported-regions page at https://www.anthropic.com/supported-countries, and SDAIA's Regulation on Personal Data Transfer Outside the Kingdom. The prospectus had not been publicly filed as of 29 September 2026, and figures and wording may change in the public version.