Artificial intelligence company Anthropic has proposed an industry-wide pause framework for advanced AI development, warning that increasingly capable systems could eventually accelerate their own development faster than humans can safely oversee. The Claude maker said AI systems are already playing a growing role in building software, conducting research and completing technical work that previously required significantly more human effort. Anthropic employees have increasingly relied on Claude for coding and other development tasks. One employee said the shift had become so extensive that they had gone months without writing code themselves. The company said fully autonomous self-improving AI systems do not yet exist, but advances in coding, research and automation could substantially accelerate future model development. Anthropic also said Claude-generated code had progressed from being somewhat worse than human-written code to roughly comparable quality at the time of its report, with the company expecting AI-generated code to eventually become consistently better. As an example, Anthropic said Claude produced more than 800 software fixes that reduced one category of API errors by roughly 1,000 times. An engineer estimated that completing the same work manually could have taken a person about four years. Anthropic warned that the bigger concern is recursive self-improvement, where increasingly capable AI systems help researchers develop even more powerful models, potentially compressing years of technological progress into much shorter periods. The company argued that frontier AI labs should establish a coordinated pause mechanism in advance. Such a framework could give researchers, governments and regulators time to evaluate risks if AI capabilities begin advancing faster than safety measures and oversight can keep pace.