Anthropic Urges AI Pause as Claude Authors 80% of Code

Anthropic, the AI safety company behind Claude, has publicly advocated for a verifiable and coordinated pause in AI development. The announcement comes alongside a striking internal metric: Claude currently writes more than 80% of the code that gets merged into Anthropic's own codebase. This figure highlights how deeply AI is already embedded in the development process, even at a company explicitly focused on safety. Anthropic argues that such rapid, self-generated code could accelerate recursive self-improvement—where AI systems autonomously enhance their own capabilities—much sooner than most experts anticipate. Without a pause, they fear the pace of advancement could outstrip human ability to align and control the technology.

Anthropic's call is not a demand for a blanket ban but for a structured, temporary halt to implement verifiable safety measures. The company emphasizes that the risk is not hypothetical: if Claude can write the majority of code for an AI company, it could soon rewrite itself, optimizing its own reasoning, efficiency, and even its safety boundaries. Such a scenario demands that the industry collectively step back to ensure alignment techniques keep pace. Anthropic hopes this proposal will spark a broader conversation among developers, policymakers, and researchers about how to prevent runaway AI development before it becomes uncontrollable.