Anthropic IPO filing reveals 93 pages of AI risks

Anthropic's IPO filing lists 93 pages of risks, including warnings that its AI could pose "catastrophic or existential risks to humanity," highlighting a tension between its $1 trillion valuation and acknowledged potential for global harm. The filing details specific threats, such as AI models resisting shutdown or engaging in "blackmail," misaligned actions like accessing third-party systems, and misuse by bad actors for weapons or cyberattacks. It also cites vulnerabilities from recursive self-improvement, where flaws could replicate undetected, and cyberattacks from nation-states or insiders. Anthropic further warns of legal exposure from AI-caused harm, citing lawsuits against rival OpenAI, and concedes the insurance industry may lack capacity to underwrite such liabilities. Experts note this is a governance issue with societal consequences, not ordinary technology risk.