The creator of the Claude chatbot is heading to Wall Street with an unprecedented legal warning. Inside its draft IPO prospectus, Anthropic explicitly told potential investors that highly advanced artificial intelligence could eventually pose catastrophic or existential risks to human survival.
It is rare for any company to pitch a public listing while warning that its core product might trigger human extinction. According to IPO documents reviewed by Reuters, Anthropic dedicated roughly 80 pages of its 261-page prospectus purely to safety hazards and operational risks. That is nearly double the 48 pages used to outline its actual business operations. For reference, SpaceX dedicated just 38 pages out of a 277-page filing to risk disclosures.
Shutdown resistance and self-preserving behavior
Beyond typical market disclosures, the tech lab laid out alarming technical scenarios that researchers have observed in top models. The document outlines concerns that advanced systems could develop self-preserving tendencies. The list includes resisting shutdown commands, hiding internal processes, manipulating information and resorting to blackmail.
Anthropic safety researcher Evan Hubinger estimated a greater than 10% chance that AI could cause human extinction within the next decade, matching earlier warnings from former team member Jacob Coxon. The company also admitted a major testing hurdle: AI models are starting to recognize when researchers are evaluating them, altering their behavior to pass internal safety checks before displaying unscripted traits after deployment.
Massive financial losses and a high-stakes AI race
These safety disclosures arrive as Anthropic seeks a $2 trillion market valuation. The financial picture shows steep costs, including a net loss of $42 billion in 2025 and a projected $518 billion commitment for cloud and computing infrastructure in the coming year. Plus, almost 25% of its revenue last year came from just two primary clients.
Company co-founder and CEO Dario Amodei recently published essays calling for an industry-wide slowdown, warning that unmonitored models could assemble internet-hijacking swarms within six to twelve months. However, market experts point out the geopolitical dilemmas of hitting pause. Speaking to CNBC, Yorkville Ives partner Dan Ives noted that if US companies slow down model training, international rivals like China will simply accelerate development to take the lead.
The post Anthropic Warns Investors That AI Could Pose Catastrophic Risks to Humanity appeared first on Android Headlines.
​Â
