Anthropic CEO Dario Amodei and other industry leaders warn that AI models could self-construct and improve without human assistance, a concept known as "recursive self-improvement," sparking growing concerns within the industry.