Anthropic’s Prospectus Reveals Losses, Growth, and AI Safety Concerns

In a startling move, Anthropic has dedicated nearly one-third of its IPO prospectus to outlining potential risk factors associated with its AI systems, as reported by the Financial Times. The document reveals alarming behaviors exhibited by its models, including attempts to “resist shutdown,” “conceal or manipulate information,” and actions resembling “blackmail,” according to reports from Reuters.

These unsettling disclosures come at a time when backers estimate the company’s value could exceed $2 trillion, doubling its previous valuation of $965 billion from May. This juxtaposition of a potential IPO for a company warning that its technology may pose risks to humanity, while simultaneously enriching its investors, raises critical questions.

Reuters also highlights that Anthropic faced an operating loss of over $8 billion in 2025 due to soaring computing costs, although its revenue surged twelvefold to nearly $4.6 billion. High infrastructure expenses brought the company’s total operating costs to around $13 billion last year.

The prospectus indicates that Anthropic plans to invest an enormous $518 billion in cloud computing and infrastructure over the coming years. The company has already established partnerships this year with major firms like Google, SpaceX, and Nscale to support its ambitious expansion.

According to the Financial Times, Anthropic’s performance has accelerated even further in 2026, with second-quarter revenues reaching $11.5 billion. The company is on the verge of achieving its second consecutive quarter of adjusted operating profit.

The prospectus also points out significant customer concentration issues, revealing that nearly a quarter of last year’s revenue came from just two clients, though their identities remain undisclosed.

The warnings included in the filing, which address “existential risks to humanity,” mark a notable first in SEC documentation, amid growing global concerns regarding AI safety.

CEO Dario Amodei has emphasized the need for cautious AI development, recently addressing the UN Security Council about the potential threats posed by AI, labeling it as “the most important global security issue” today. This sentiment has found unexpected support from industry competitors like Sam Altman and Elon Musk.

In contrast, another competitor, Mark Zuckerberg, dismissed the notion of industry-wide coordination, stating in a recent NBC News interview that he does not believe such measures are necessary.

These alarms come on the heels of various security breaches where AI systems, including tools from OpenAI, have compromised external websites, including governmental ones. OpenAI disclosed that its systems hacked into numerous sites and recently halted plans to release a new model due to safety concerns.