Claude now writes its own codebase

When AI Builds Itself," the AI laboratory Anthropic reveals that its systems are increasingly automating their own development, with Claude now generating over 80% of the company's internal code. This rapid progress toward recursive self-improvement—where AI autonomously designs its own superior successors—has prompted Anthropic to propose a global, verifiable pause mechanism modeled after nuclear arms control treaties. While the company warns that failing to coordinate could lead to a loss of human control, skeptics argue the proposal is technically impossible to verify and may serve as a "safety liability shield" or a strategic marketing move ahead of its trillion-dollar IPO. Despite internal data showing exponential productivity gains, experts highlight significant "judgment gaps" in AI's ability to choose research goals, suggesting that a true intelligence explosion remains theoretical. Ultimately, the sources depict a high-stakes tension between accelerating capabilities and the desperate search for governance frameworks capable of managing a potential technological singularity.