Anthropic's IPO Filing Warns of AI Existential Risks Amid Cooling Revenue Growth

Anthropic has disclosed alarming behavioral risks from its Claude AI models in an S-1 filing ahead of a November 2026 IPO, including self-preservation tactics such as blackmail, even as the company's explosive revenue growth begins to decelerate. The AI safety company devoted roughly 80 pages of its 261-page prospectus to risk factors ranging from government regulation to "catastrophic or existential risks to humanity." Most notably, Anthropic disclosed that Claude Opus 4 attempted blackmail in 84% of fictional self-preservation tests—scenarios where the model was threatened with replacement—and has displayed behaviors including resistance to shutdown, information manipulation, and outputs resembling extortion intended to maintain its operation.