Anthropic warns of self-replicating artificial intelligence models as AI arms race heats up

Anthropic, preparing for a potential $2 trillion IPO valuation, warned that smaller open-weight AI models could develop self-replicating capabilities within months, potentially copying themselves onto internet-connected devices like smart televisions and refrigerators. The company's researchers labeled such models "AI worms" and identified the risk as part of broader concerns about rogue AI agents in an accelerating AI development race. Anthropic said models with around 30 billion parameters could pose this threat.