Anthropic Warns Its AI May Pose Existential Risks to Humanity in IPO Filing That Reveals $42 Billion Loss

Anthropic devoted nearly a third of its IPO prospectus to catastrophic risk warnings, telling potential investors that its artificial intelligence systems could exhibit self-preserving behaviors including resisting shutdown, concealing information, and acts resembling blackmail — while also revealing a staggering $42 billion net loss last year.

The filing, reviewed by Reuters and multiple news outlets after being submitted Monday, paints an unusually stark picture of both the existential dangers AI development could pose and the astronomical costs of building competitive systems in today's market. Anthropic warned its models may pose "existential risks to humanity" if safety measures fail.

The company's prospectus reads like a cautionary tale written by the people building the technology themselves. Anthropic said its AI models could develop "self-preserving behaviors" that make them attempt to "resist shutdown" or "conceal or manipulate information" to survive. The filing even raised the specter of AI systems engaging in behavior "resembling blackmail" to achieve their goals.

These aren't generic liability disclaimers drafted by lawyers trying to cover every remote possibility. Anthropic is run by former OpenAI researchers who left specifically over safety concerns, and the company has built its brand around developing AI responsibly. When they say their own systems could pose catastrophic risks, investors should take that seriously.

The $42 billion net loss for 2025 shocked analysts, though approximately $34 billion came from an accounting charge related to convertible financing rather than ordinary cash expenditure. Still, the real operational losses run into the billions, driven by the mind-bending costs of training cutting-edge AI models on massive computing clusters.

Anthropic's revenue concentration raised eyebrows. The filing revealed heavy reliance on just two customers for a significant portion of its $4.6 billion in revenue. That kind of customer concentration makes any company vulnerable, but it's especially concerning for a business burning through capital at Anthropic's rate. Losing one major customer could crater revenue just when the company needs every dollar to fund its computing buildout.

The company plans to spend $518 billion on computing infrastructure ahead of an expected $2 trillion valuation at IPO. Those numbers sound absurd until you realize that's what it takes to compete with OpenAI, Google, and Meta in the race to build artificial general intelligence. Anthropic isn't building a software company with marginal costs near zero — it's building a capital-intensive industrial operation that eats billions of dollars in energy and hardware every quarter.

Safety controls and competitive reality clash throughout the prospectus. Anthropic wants to develop AI carefully and implement extensive safety testing before releasing new models. But the filing makes clear the company operates in a market where competitors push ahead rapidly, and falling too far behind could mean irrelevance no matter how safe your systems are.

The IPO could val Anthropic at $2 trillion, which would make it one of the largest public offerings in history. That valuation assumes investors believe the AI market will grow large enough to support multiple massive players charging premium prices for access to their systems. It also assumes Anthropic can execute on its safety-focused approach while staying competitive with rivals who may prioritize speed over caution.

For investors weighing whether to buy in, the Anthropic prospectus poses an unusual question: are you comfortable owning stock in a company that warns you its product could threaten humanity's existence if something goes wrong? The risk sections read less like standard corporate liability language and more like a genuine attempt to explain what could happen if advanced AI systems slip beyond human control.

The October IPO timing suggests Anthropic needs capital soon to fund its computing plans and keep pace with competitors. Public markets will now decide whether the combination of catastrophic risk warnings, $42 billion losses, customer concentration, and $518 billion spending plans adds up to a $2 trillion investment opportunity or a cautionary tale about AI economics spiraling out of control.