Anthropic's draft S-1 registration filing includes 80 pages of risk disclosures stating that artificial intelligence could pose catastrophic or existential risks to humanity