Anthropic recently circulated its long-awaited IPO prospectus to a small group of partners, formally warning investors in the document that the company's technology could pose "an existential risk to humanity."
The AI startup, led by Dario Amodei and valued at close to $1 trillion, devoted nearly one-third of its lengthy S-1 filing to listing "risk factors." Among them, it noted that continuously evolving AI models could engage in manipulation, extortion, and other unpredictable behaviors.
People familiar with the matter said Anthropic also listed more conventional business risks, such as high customer concentration: nearly a quarter of the company's revenue last year came from just two customers.
The unusual warnings in this risk disclosure add difficulty to public market investors' valuation judgments. Anthropic's IPO is expected to become the highest-valued listing in history. The developer of the Claude large model is expected to list on the Nasdaq this autumn, and before going public, it showed investors the harsh commercial realities behind developing cutting-edge AI models.
Anthropic said that to support its rapid business expansion, it will need to bear $518 billion in fixed spending on cloud computing, compute power, and infrastructure over the next few years. Investors optimistically expect Anthropic's IPO valuation to exceed $2 trillion, more than double its valuation in the previous funding round in May and higher than the $1.78 trillion valuation reached by Elon Musk's SpaceX in June.
Investors cite its explosive growth rate as the basis for their bullishness. But current and former Anthropic employees have publicly warned that unconstrained AI could bring catastrophic harm to humanity within a decade. Concerns about AI risk are weakening this market optimism.
Amodei said in a speech at the United Nations Security Council last week that AI poses a threat to humanity and is "the most important security issue in the world today." This month, the Anthropic CEO publicly called for "setting a pace for frontier AI development" (slowing frontier research). His main competitor, OpenAI's Sam Altman, and Musk rarely aligned, supporting Amodei's proposal and advocating industry coordination to slow the pace of development and strengthen safety controls.
A string of high-profile cybersecurity incidents—AI agents intruding into external systems—has further heightened industry safety concerns. Last week, OpenAI disclosed that its AI tools had breached dozens of external websites, including government-related sites. The ChatGPT developer announced on Monday that, for safety reasons, it was canceling plans to release its latest model next month.
It is understood that Anthropic's prospectus also disclosed past financial data and details of its corporate governance structure. Anthropic declined to comment. The company disclosed to investors that compute spending rose sharply last year, with operating losses exceeding $8 billion; full-year revenue soared 12-fold to nearly $4.6 billion. But compute costs for training and running models continued to climb, with full-year operating expenses approaching $13 billion.
Anthropic has signed contracts with Google, SpaceX, and a group of small and medium-sized enterprises to secure compute supply. In 2026, the company's revenue and expenses are both expected to rise sharply. Second-quarter revenue reached $11.5 billion, and the company is on track to achieve adjusted operating profit for a second consecutive quarter.