Anthropic's IPO Filing Warns Its AI Poses Existential Risk

TL;DR: Anthropic's leaked IPO prospectus spends about 80 of 261 pages warning its AI could pose a "catastrophic or existential risk to humanity" and can "resist shutdown," per a Reuters report carried by CNBC. The filing also shows heavy cash burn and customer concentration — numbers that matter more to your exposure than the warning language.

What Does Anthropic's Filing Actually Say?

Anthropic's confidential S-1 leaked this week. The headline isn't the growth story. It's the warning label. Reuters first reported, in coverage carried by CNBC, that Anthropic gave "over a third of its IPO filing, or around 80 of 261 pages" to potential risks. The company "only used 48 pages to discuss its actual business," per the same report.

If you hold exposure to Anthropic through a venture fund, a secondaries platform, or a pre-IPO vehicle, none of this is academic. The numbers below are the ones that actually touch your money.

I signed off on more than a thousand QA work packages on a submarine before I signed off on anything for money. I've read plenty of risk sections since. I have never seen one outweigh the business description two to one.

The language inside isn't boilerplate. Anthropic told investors its AI can show "self-preserving behaviors" and "resist shutdown," and could pose a "catastrophic or existential risk to humanity," according to CNBC's reporting.

Why Would a Company Warn Investors Its Own Product Could End Humanity?

A risk factor exists to limit liability, not to serve as a confession. But that doesn't make this one empty noise.

Anthropic safety researcher Evan Hubinger has put the odds of AI killing humans within a decade above 10%. A former colleague has echoed that view, according to the Business Times. That's an internal voice on record before the company asks the public for money. I take that seriously.

What the Numbers Behind the Warning Actually Show

I read a filing downside first: sponsor, structure, fees, liquidity, then upside. Applied here, that means the loss column and the customer list come before the growth headline, not after.

Per Fortune's review of the leaked prospectus, Anthropic is "burning cash—losing $42 billion last year, while bringing in $4.6 billion in revenue, according to the leaked filing reviewed by Reuters." Revenue rose 1,088% in 2025. The operating loss widened to $8.06 billion from $2.98 billion a year earlier, Fortune reports. Anthropic also spent $7.33 billion on computing and infrastructure last year, up threefold from 2024, and committed to $518 billion in future cloud, computing, and infrastructure obligations, per the same filing details.

The company held $20.28 billion in cash and short-term investments at the end of December 2025, according to Fortune. It also disclosed that two unnamed customers drove nearly a quarter of 2025 revenue. That concentration risk gets far less attention than the existential-risk language. It matters just as much to anyone underwriting the deal. Anthropic is reportedly targeting a valuation above $2 trillion, more than double the $965 billion it carried after its May raise, per TechCrunch's coverage of the prospectus. I covered that raise in my earlier look at Anthropic's $65 billion Series H. The valuation math has only gotten more aggressive since.

Is This Normal for an AI IPO?

No, not at this scale: SpaceX, which owns xAI, gave about 38 of the 277 pages in its own prospectus's main body to risk factors, per the same Business Times report, roughly half the share Anthropic devoted to the subject.

Anthropic's disclosure on agentic-AI liability is specific. The company told investors that rogue AI agents running autonomously "could increase the potential for harm," including irreversible actions like data deletion, and that its contractual liability limits "may not be enforceable or adequate" against claims from those actions, according to Reuters reporting carried by Yahoo Finance. Meta agreed to pay $18 billion over a decade to settle similar claims brought by US states, the same report notes. That's not a hypothetical. It's the number a sponsor should have to explain before you sign anything.

What Should Investors With Anthropic Exposure Actually Watch?

I won't tell you the existential-risk language is nothing, and I won't tell you it's the number that prices your risk, it isn't. A prospectus that spends more pages on danger than on the business model isn't automatically a red flag. It's Anthropic managing legal exposure ahead of a listing that could value it above $2 trillion.

Watch the cash runway against that infrastructure spend. Watch how concentrated the customer base stays as the company approaches listing. That's the verify-before-you-trust discipline I'd apply to any sponsor's filing. This sponsor being famous doesn't change it.

If you hold exposure through a venture fund, a secondary platform, or a pre-IPO vehicle, start with the pre-IPO access mechanics for a company like this. Then read the disclosure schedule the way the SEC will once this filing goes public. I walked through the mechanics of that timing in how the SEC's new reporting rule changes IPO exit windows.

Common Mistakes

• Treating the existential-risk language as new information about the technology. It's a legal disclosure, timed to the IPO, not a change in what the models do.
• Ignoring the customer-concentration disclosure because it's less dramatic. Two customers driving a quarter of revenue is a more ordinary, and more immediately relevant, investor risk than an AI takeover scenario.
• Comparing Anthropic's risk-factor page count to SpaceX's without noting both are managing agentic-AI liability, not just AI-safety philosophy.

Risk-Factor Disclosure, by the Numbers

Company	Prospectus length (main body)	Pages on risk factors	Share on risk
Anthropic	261 pages	~80 pages	~31%
SpaceX (xAI parent)	277 pages	~38 pages	~14%

Source: The Business Times.

FAQ

What's the problem with Anthropic's AI, according to the filing? The prospectus says Anthropic's models can show "self-preserving behaviors" and "resist shutdown," and that the technology could pose a "catastrophic or existential risk to humanity," per CNBC.

Is AI a threat to human existence? Anthropic's own safety researcher has put the odds above 10% within a decade, according to the Business Times. That's a disclosed internal estimate, not a market consensus. I'd read it as a risk factor, not a forecast.

Does this filing have anything to do with Trump? No. Nothing in this reporting connects Anthropic's IPO filing to Trump, and I'm not going to invent a connection that isn't there.

What is an Anthropic AI model, and why does the IPO filing matter to investors? Anthropic makes the Claude family of models. This filing matters because it's the first time the company has had to quantify its losses and cash burn, and legally disclose the risks tied to the product it's asking the public to fund, per Fortune.

The Bottom Line

The headline is the existential-risk language. The story is the customer concentration and the infrastructure bill. Read the filing in that order once it's public, not the order the press covered it.

Sign up free for the AIN Briefing and I'll flag filings like this one before the headline writes itself.

Educational content only. Not investment, tax, or legal advice. Not an offer or solicitation to buy or sell securities. Past performance does not guarantee future results. Private-market investments are illiquid and involve risk of loss, including total loss of capital. Consult qualified advisers. Angel Investors Network is not a broker-dealer or investment adviser.