Anthropic Advocates for Global Pause on AI Development to Mitigate Risks
- Over 80% of code in Anthropic’s internal codebase was generated by AI system Claude by May 2026, up from a few percent before February 2025.
- Average engineers at Anthropic integrated eight times more code per day in Q2 of the same year compared to the previous year.
- Claude Mythos Preview autonomously completed tasks lasting over 16 hours and fixed more than 800 software bugs, tasks that typically take years for humans.
- Anthropic calls for international coordination to manage risks associated with AI’s rapid advancement and recursive self-improvement.
- The proposal comes as Anthropic prepares for an IPO, underscoring the urgency of addressing potential challenges posed by advanced AI systems.
Anthropic emphasizes the need for a mechanism to temporarily halt advanced AI development when serious risks arise, highlighting the increasing influence of AI in coding and research activities. This call aims to ensure global stability amid rapidly evolving technologies that could outpace regulatory frameworks.
With over 80% of its codebase written by Claude, Anthropic’s push for a global pause reflects significant concerns about losing control over advanced AI systems.(Source)
