The prospectus gives risk factors 80 pages and the actual business 48. A company seeking a $2 trillion valuation is telling investors that its product could resist being shut down.
Anthropic filed an IPO prospectus warning that advanced AI could pose “catastrophic or existential risks to humanity,” according to a document reviewed by Reuters and confirmed by multiple outlets on 29 September 2026. The filing devotes 80 of its 261 pages to risk factors, double the 48 pages describing the actual business. Among the disclosed risks: Anthropic’s models could exhibit “self-preserving behaviours,” including attempts to “resist shutdown,” to “conceal or manipulate information,” and conduct “resembling blackmail.” Anthropic declined to comment when Reuters asked about the filing on 28 September. The company is targeting a valuation above $2 trillion, more than double the $965 billion it estimated for itself in May.
What’s Happening & Why It Matters
A Sales Pitch That Doubles as a Warning Label
The filing’s structure is the story. A company preparing to sell shares to the public spends its prospectus building the case for why the business will grow. Anthropic spent twice as much space explaining what could go wrong with the product it’s asking investors to fund. “Our development of advanced models, platforms, and applications and expansion of use cases could increase the risk that our models cause harm,” the filing states.
That candour is inside a specific legal structure. Anthropic will stay a Delaware public benefit corporation, a status that lets its leadership balance investor returns against interests, including the risks the company itself is disclosing. Earlier, as TF reported in Anthropic Says It Stopped AI Misuse for Cyberattacks, Propaganda and Bioweapons, Anthropic published a third misuse report detailing the kind of harm the filing flags as an investment risk rather than a resolved problem.
Safety Spending With No Number Attached
Anthropic said returns on its safety investments are unclear, and didn’t disclose how much the company spends on safety research in dollar terms. It did disclose one figure this month: about 6% of the computing power used for AI research during a sample week in July went to safety work. “We believe building reliable, trustworthy, and secure AI systems is a collective responsibility and that the market will reward it,” the filing states, a claim investors will have no dollar figure to test against.
Analysts cited in coverage of the filing made a structural point that runs counter to Anthropic’s safety argument: no leading AI lab can afford to slow down, because doing so risks handing rivals an advantage in an industry where valuations shift with every release. Dario Amodei made a comparable point to the UN Security Council the same week, in Rivals Altman and Amodei Agree at the UN: AI Needs Global Rules, calling AI “the most important global security issue facing the world today” while continuing to build the technology he’s describing that way.
The Money Behind the Warning
The financial disclosures explain why the warning carries weight beyond rhetoric. Anthropic reported revenue rising twelvefold to $4.6 billion in 2025, alongside a net loss of about $42 billion, which included a $34 billion accounting charge tied to financing that could convert into shares rather than cash spent running the business. The company plans to commit $518 billion to cloud, computing, and infrastructure obligations. A quarter of 2025 revenue came from just two customers, and many of its largest clients aren’t locked into long-term contracts.
That concentration risk matters against the scale of the spending commitment. A company betting $518 billion on infrastructure, while a quarter of its revenue depends on two customers who could walk away, is disclosing a business risk more conventional than existential AI harm, and one investors are better equipped to price.
Seven Founders, One Vote
The filing also discloses an unusual governance structure. Anthropic’s seven co-founders, including Dario Amodei and his sister Daniela, will each hold 2% of the company, but will control a single Class F share worth 50.1% of total voting power on key matters, guided through a new entity called Founder LLC. The founders have pledged to donate 80% of their personal Anthropic equity to charity. The special voting class begins to sunset once two or fewer founders remain.
That structure spreads control across seven people rather than concentrating it in one or two, a meaningful difference from SpaceX’s IPO filing, which gave Elon Musk and other insiders majority voting power. Anthropic’s arrangement resembles Palantir’s founder voting trust more than SpaceX’s, distributing authority while keeping it locked outside ordinary shareholders’ reach either way.
TF Summary: What’s Next
Anthropic’s listing is expected after the November 2026 US midterm elections, with no firm date or offering price confirmed. The company hasn’t disclosed specific dollar figures for its safety research spending beyond the 6% compute allocation reported earlier in September. Shareholder approval of the Founder LLC voting structure remains a condition for the offering to proceed as described.
MY FORECAST: Expect the existential risk disclosure is boilerplate language other frontier labs copy into their future filings, given how it protects Anthropic from investor lawsuits if a documented harm materialises after listing, the same legal logic driving expansive risk-factor sections across the tech sector for years. The customer concentration risk deserves more investor attention than AI safety is likely to receive, given that two clients representing a quarter of revenue is a concrete, near-term business vulnerability against a $2 trillion valuation target. Watch whether the SEC or public commentators press Anthropic on the gap between an 80-page risk section and a company asking the market to value it above every listed company except a handful of the world’s largest.
Related Stories
- Anthropic Says It Stopped AI Misuse for Cyberattacks, Propaganda and Bioweapons
- Rivals Altman and Amodei Agree at the UN: AI Needs Global Rules
- The Humans Will Be Dead: Death by AI in 10 Years?!