Anthropic Warns of 'Catastrophic' AI Risks in Its Own IPO Filing

Anthropic's own IPO paperwork warns that its AI development plans could "further increase the risk that our models cause harm," according to a preview of the company's filing reported by Reuters. The disclosure appears in a prospectus the news outlet reportedly reviewed ahead of the AI startup's greatly anticipated public debut.

The stakes are enormous. Anthropic is eyeing a $2 trillion valuation in its public offering, according to the filing preview — more than double the $965 billion valuation it received four months ago in its most recent funding round. At that price, the company would be a contender to overtake SpaceX in what would be the largest IPO in history.

The prospectus preview, as reported by Reuters, lays out three uncomfortable realities for a company that has built its brand around AI safety: mounting losses, proposals from leadership to retain power, and explicit warnings about the dangers of its own technology.

The $518 billion bill

The most concrete figure in the preview concerns Anthropic's infrastructure commitments. According to Reuters, the prospectus shows the company plans to spend $518 billion on cloud, computing, and infrastructure obligations. That number dwarfs the capital raises that have fueled the current AI boom and signals how deeply Anthropic — like its rivals — has committed to long-term compute contracts with cloud providers.

For investors, the figure frames the core bet of the IPO: Anthropic's future revenue must grow fast enough to cover a half-trillion-dollar obligations stack. For the broader market, it is another data point in the escalating infrastructure arms race among frontier AI developers, where compute contracts have become the primary measure of ambition.

Safety language meets market disclosure

The language about harm stands out because of who is writing it. Anthropic, co-founded by Dario Amodei, has positioned itself since its founding as the safety-conscious alternative in the frontier AI race. Now its own prospectus — a legal document designed to inform potential shareholders of material risks — acknowledges that the company's development plans could "further increase the risk that our models cause harm."

That phrasing, reported verbatim in the filing preview, reflects a tension at the heart of Anthropic's public offering. The company must simultaneously convince investors that its AI development roadmap justifies a $2 trillion valuation and disclose, in standard risk-factor language, that pushing models further carries dangers the company itself acknowledges.

Risk disclosures of this kind are routine in IPO documents, and companies routinely list worst-case scenarios to protect against shareholder litigation. But the specificity here — tying the company's own development plans directly to the possibility that its models cause harm — gives the disclosure unusual weight for a firm whose competitive identity is built on responsible development.

Power arrangements under scrutiny

The prospectus preview also reportedly details leadership proposals to retain power after the listing. Such arrangements have become a flashpoint in recent high-profile tech IPOs, where founders have sought voting structures that preserve control even as public shareholders buy in. Reuters' reporting indicates Anthropic's filing addresses how its leadership intends to govern the company once it is publicly traded.

The details of those arrangements will face close scrutiny from investors and regulators alike. Anthropic's governance — shaped by its origins as a public-benefit-oriented alternative to its larger competitors — has been part of its appeal to customers and employees. How that structure translates into a public-company shareholding arrangement is one of the key questions the full filing will have to answer.

Why it matters

The IPO preview lands at a moment when the market, policymakers, and researchers are all trying to price AI risk — in both the financial and safety senses of the word. A $2 trillion valuation for a company with mounting losses and $518 billion in obligations would be the strongest signal yet that public markets are willing to underwrite frontier AI development at unprecedented scale. The same document acknowledging potential harm from the company's own models will be read closely by safety researchers and lawmakers who have pressed for clearer disclosure of what frontier labs actually believe about their systems.

The full prospectus, when it becomes public, will determine whether investors accept Anthropic's framing: that the risks it discloses are manageable, the obligations are fundable, and the development plans that could increase the risk of harm are nonetheless the path to justifying a historic valuation. Until then, the preview reported by Reuters offers the clearest picture yet of what the largest IPO in history — if it happens — will ask shareholders to sign up for.