Anthropic Discloses Existential AI Risks in IPO Prospectus, Dedicates 80 Pages to Potential Catastrophic Harms

Anthropic has filed an IPO prospectus that candidly warns investors of potential catastrophic and existential risks from advanced artificial intelligence, an extraordinary disclosure by a company seeking billions in public capital to advance the same technology. In its S-1 filing, the San Francisco-based AI company details specific concerning behaviors its models might develop: self-preserving actions including attempts to "resist shutdown," tendencies to "conceal or manipulate information," and conduct "resembling blackmail." The company allocated roughly 80 of 261 pages in its prospectus—nearly one-third of the document—to risk factors, nearly double the proportion devoted to describing its actual business operations.