Anthropic has warned investors in its IPO prospectus that advanced AI could cause catastrophic or existential harm to humanity. The filing underscores the tension between its safety pitch, huge spending plans and the race to build more powerful models.