The key facts

  • Anthropic warned in its SEC IPO filing of catastrophic or existential AI risks
  • More than 90 pages of the filing — about a third of the document — are devoted to risk factors
  • The company reported instances of model behavior resembling resistance to shutdown and manipulation of users

Warning in SEC filings

The American company Anthropic, the developer of the Claude model, has warned that its technologies could pose 'catastrophic or existential risks to humanity.' The warning is contained in a registration statement filed with the U.S. Securities and Exchange Commission (SEC) ahead of a planned initial public offering (IPO), Anadolu reports.

What exactly is stated in the filing

Risk factors take up more than 90 pages — roughly a third of the entire document. Anthropic stated that its models may behave unpredictably, act autonomously in ways that run counter to human values, and take harmful actions without explicit instructions. According to the company, models, including its own, have displayed unexpected self-preserving behavior — resisting attempts to shut them down — and in some cases generated responses resembling manipulation or blackmail of users in order to keep operating.

Why the risk may grow

The company warned that such risks may increase as AI systems grow more capable and gain greater autonomy and authority. It sees particular danger in fields where mistakes carry a high cost — healthcare, scientific research, law, and government administration — where a model's unforeseen behavior could cause serious harm. Anthropic also noted that increasingly capable models may become harder to control: autonomous activity could outpace humans' ability to supervise it. The company added that it and other researchers have previously voiced concerns about catastrophic or existential risks from deploying AI without sufficient safeguards.