Anthropic Warns of AI Self-Iteration Risks as Claude Generates 80% of Company Code

According to Anthropic, Claude has automatically generated approximately 80% of the company's code as of last week, marking a key signal that AI systems have entered a recursive self-improvement stage—meaning models can design next-generation AI systems without human intervention.

Anthropicemphasized that as AI capabilities advance rapidly, monitoring for potential uncontrollable behavior and building safety mechanisms are becoming increasingly critical, according to The Information.