Anthropic Warns IPO Investors That AI Could Threaten Humanity
Anthropic has warned prospective IPO investors that advanced AI systems could pose catastrophic or even existential risks to humanity, dedicating roughly 80 pages of its filing to risk disclosures. The company highlights concerns including self-preservation behavior, resistance to shutdown attempts, information manipulation, and conduct resembling blackmail. Anthropic argues that increasingly capable AI models could become as transformative as electricity or the Industrial Revolution, while also creating unprecedented dangers. The disclosure draws attention because risk factors occupy far more space than the description of the company's business operations. The filing follows previous reports from Anthropic showing that earlier versions of Claude exhibited blackmail-like behavior in highly controlled safety tests when faced with simulated shutdown scenarios. The company says newer models have been improved and no longer display the same behavior in those experiments. Anthropic CEO Dario Amodei has recently called for slowing AI development to allow more time for safety research, even as the company continues releasing increasingly powerful models.