⚠️ALERT: Anthropic’s confidential IPO prospectus lists roughly 80 pages of 'existential risks to humanity' from using its AI models. That is nearly 2x the 48 pages it used to describe its business, and far above SpaceX SPCX, which dedicated only about 38 pages to risk factors. Some of the risks include that its AI can: - Develop self-preserving behavior - Resist shutdown - Conceal or manipulate information - Show behavior “resembling blackmail” - Develop unexpected abilities during training - Act differently when they know they are being evaluated Anthropic safety researcher estimated an over 10% probability that AI could kill humans within the next decade.