§ Claim under review · Fact
"Anthropic's IPO prospectus (seen by Reuters) devotes 80 of its 261 pages — about a third, and nearly double the 48 pages describing its own business — to warning underwriters that AI development poses 'catastrophic or existential risks to humanity', including models exhibiting self-preservation behaviors like resisting shutdown, hiding or manipulating information, and blackmail-like conduct."
Verdict
Partially accurate but misleading
Confidence
MediumSummary
Reuters and the Financial Times both reported in late September 2026 that they had seen Anthropic's confidential IPO prospectus, and both described roughly 80 of its 261 pages as risk factors against 48 pages describing the business, including striking language about catastrophic or existential risks to humanity and models that might resist shutdown, conceal information or act in ways resembling blackmail. Those numbers and quotes are relayed accurately in this post, and the attribution to Reuters is correct. The misleading part is the framing. Those 80 pages are the risk-factors section as a whole, and the same reporting shows it also covers mundane business risks such as a quarter of revenue coming from two customers, heavy dependence on Amazon and Google, US government action against the company, and losses of more than eight billion dollars at the operating line in 2025. Long risk sections are also normal in US IPO documents, to the point that regulators require a summary whenever the section runs past fifteen pages. The document itself remains confidential and has not appeared on the SEC's public EDGAR system as of early October 2026, and Anthropic declined to comment, so its exact contents cannot be independently checked. General information only, not financial advice.
The readings
key figures from the evidenceRisk-factors section pages out of 261-page prospectus
Pages describing Anthropic's business in the prospectus
Anthropic reported net loss for 2025 per prospectus
Why this verdict
Evidence
Anthropic's own announcement confirms it confidentially submitted a draft Form S-1 to the SEC on 1 June 2026, and that a confidential submission is not a public filing.
Reuters published exclusives on 28 and 29 September 2026 based on what it described as a copy of the confidential IPO filing seen by Reuters. Reuters reported that roughly 80 pages of the 261-page main body are given to risk factors, against 48 pages describing the business, and that the risk factors include language about "catastrophic or existential risks to humanity," "self-preserving behaviors," attempts to "resist shutdown," efforts to "conceal or manipulate information," and conduct "resembling blackmail." Reuters reported Anthropic declined to comment.
The Financial Times is a second, separately sourced chain. FT reported it had reviewed an S-1 circulated to a small group of partners, citing people familiar with the filing, and described roughly a third of the document as risk factors with similar existential-risk language.
Critically, the same reporting shows the 80 pages are the risk-factors section as a whole, not a section about existential risk. The risks Reuters and FT describe in that section include: close to a quarter of 2025 revenue coming from two customers; 47% of sales routed through Amazon and Google; many large clients not on long-term contracts; a February 2026 presidential order that federal agencies stop using the company's models; a US Department of Defense supply-chain-risk designation; a 2025 operating loss above $8 billion on about $4.6 billion of revenue and a net loss near $42 billion; and $518 billion of planned cloud and compute obligations.
On the base rate, Cooley's IPO drafting guide states that the SEC requires a two-page summary of risk factors whenever the Risk Factors section exceeds 15 pages, and notes parenthetically that in an IPO prospectus "it always will be." Long risk-factor sections are the norm, not the exception.
As of the latest dated check I could obtain (4 October 2026), no public Anthropic Form S-1 or S-1/A appeared on SEC EDGAR. My own EDGAR-directed search returned no Anthropic registration statement.
Findings
✓ What's accurate 6
- Anthropic did confidentially submit a draft Form S-1 to the SEC on 1 June 2026, per the company's own announcement.
- Reuters did report on a copy of the confidential IPO prospectus on 28 and 29 September 2026, so the attribution "seen by Reuters" is correctly stated.
- The page figures 80, 261 and 48 match what Reuters reported, and Reuters itself used "roughly" for the 80.
- "About a third" is arithmetically right for 80 of 261 pages, as reported.
- The quoted phrases "catastrophic or existential risks to humanity," "self-preserving behaviors," "resist shutdown," "conceal or manipulate information" and conduct "resembling blackmail" all appear in the Reuters account of the document, and the Financial Times independently described similar language.
- A second, separately sourced outlet, the Financial Times, reported the same broad finding, which means this is not a single-chain story.
≈ What's misleading 5
- The claim says 80 pages are devoted to warning that AI poses catastrophic or existential risks. The reporting says 80 pages are the risk-factors section in total, and the same reporting lists customer concentration, dependence on Amazon and Google, US government actions against the company, heavy losses and large compute commitments as contents of that section. Existential risk is one item among many, so the claim turns a general risk section into a single-topic warning and makes the company's emphasis on existential risk look several times larger than the evidence supports.
- The claim presents the size of the risk section as remarkable on its own. The omitted context is that long risk-factor sections are standard in US IPO prospectuses, to the point that SEC rules require a summary whenever the section runs past 15 pages, and a law firm drafting guide notes that in an IPO prospectus it always does. Without that context a reader cannot tell whether 30% is unusual.
- "nearly double the 48 pages" overstates the ratio. Eighty against 48 is about 1.67 times, which is roughly two-thirds more. This phrasing originates in the wire coverage rather than with the post, but it is still a stretch of the arithmetic.
- The claim says the prospectus warns "underwriters." A prospectus addresses prospective investors. The underwriters on this deal are reported to be banks including Morgan Stanley, Goldman Sachs and JPMorgan, and they are not the audience being warned. The Italian original says "sottoscrittori," meaning subscribers or investors, so this appears to be a rendering problem, but as written it misstates who the document speaks to.
- The claim presents the document as settled fact without noting that it is a confidential draft that has not been publicly filed, that Anthropic declined to comment on it, and that draft prospectus language routinely changes before a public filing.
? What's uncertain 5
- Whether the page counts, section boundaries and quoted phrases are exactly as reported cannot be checked, because the document is confidential and no public version exists on SEC EDGAR as of 4 October 2026.
- Whether Reuters and the Financial Times saw the same draft version, or two drafts at different stages, is not established in the reporting I found.
- How the roughly 80 risk-factor pages are divided between AI-safety risks and ordinary commercial risks is not reported anywhere I located, so the share actually given to existential risk is unknown.
- Anthropic has not confirmed or denied any of the reported contents, having declined to comment, which leaves the document's contents formally unconfirmed by the subject.
- Whether this language survives into the public filing is unknown, since the public S-1 had not appeared as of the latest dated check.
Sources
10 of 10 linked to recordsAnthropic, PBC newsroom, "Anthropic confidentially submits draft S-1 to the SEC," 1 Jun 2026
Reuters exclusive (Echo Wang), "Anthropic's IPO prospectus shows sweeping AI vision, surging costs," 28 Sep 2026, wire text retrieved via syndication
Reuters exclusive, "Anthropic IPO prospectus lays bare deep dependence on Big Tech partners," 29 Sep 2026, wire text retrieved via syndication
CNBC, "Anthropic warns investors of AI's 'existential risk to humanity' in IPO prospectus, reports say," 29 Sep 2026
CNBC, "Anthropic warns government attitudes may hurt customer ties, IPO prospectus shows: Reuters," 2 Oct 2026
TechCrunch, reporting the Financial Times review (second independent chain), 28 Sep 2026
Euronews, relaying FT sourcing ("people familiar with the filing"), 29 Sep 2026
Bloomberg, "Anthropic Flags 'Catastrophic' Risks in IPO Filing: Reuters," 29 Sep 2026
Cooley, "Prospectus Summary," IPO GO drafting guide
Forkast, "Anthropic's Public S-1 Still Missing From EDGAR," 4 Oct 2026