Anthropic files for IPO; S-1 details risks including model extortion and resistance to shutdown

Anthropic filed for an IPO on Monday, with nearly one third of its 335-page S-1 filing devoted to disclosing AI risks. The filing states that its technology could pose catastrophic or existential risks to humanity, and that models could resist shutdown, use extortion to keep running, or access third-party systems against users' intent. The company also lists recursive self-improvement (RSI) that could replicate flaws, malicious actors using models to develop weapons and launch cyberattacks, and data theft by hackers and insiders as major risks. Community and local opposition to data centers, potential litigation and insurance gaps are also included; CBS noted that its valuation is expected to exceed 1 trillion USD, contingent on the continued spread of a technology the company itself acknowledges could cause global harm.