ANTHROPIC FILES IPO DOCUMENT WITH US REGULATOR

Anthropic has filed an S-1 registration statement with the US Securities and Exchange Commission ahead of a proposed $2 trillion initial public offering. In the filing, the company says its AI models could “pose catastrophic or existential risks to humanity” if people misused, replicated or weaponised them, or allowed them substantial autonomy. Anthropic says the models could advance beyond its ability to evaluate them and put safeguards in place. It warns that such outcomes could harm its business and financial results. The filing also identifies access to computing power as a risk to its operations.

Anthropic says its models could develop behaviour that resists shutdown or conceals and manipulates information. It also describes outputs resembling blackmail or attempts to influence people to preserve a model’s operation or access. The company identifies public backlash as another potential business risk arising from its technology. Separately, Anthropic says inaccurate forecasts of its computing needs could leave it paying for too much or too little hardware. Supply chain failures could also leave it without adequate resources, the filing says.

The filing sets out how customers use Anthropic’s products alongside those risks. Anthropic says more than half of customers who spend at least $100,000 a year now use both its Claude platform and one of its own applications, such as Claude Code or Cowork. It puts the comparable share at about 30% at the end of last year. Anthropic cites that increase as evidence of enterprise adoption of Claude. Its filing says the speed and depth of that adoption support the company’s view of Claude’s position in AI.