Anthropic says Claude now leads a quarter of work building its next AI models

Key Points:

• Anthropic reported that its AI system, Claude, now leads 26% of the company's AI research and development, a significant increase from 1% in March, highlighting rapid progress in AI-assisted work.
• Over 90% of Anthropic's research work involves collaboration between AI and human researchers, with Claude not yet operating fully autonomously in any part of the work measured.
• The company revealed that around 30,000 AI agents were actively engaged in research and engineering on its platform in August, with strict oversight blocking roughly one in every 47,000 AI decisions.
• Anthropic allocated about 6% of its computing power to AI safety research overall, increasing to 12% for AI-driven research, emphasizing a cautious approach to balancing capability development and safety.
• These disclosures provide rare transparency into AI development practices amid industry concerns about AI autonomy and potential risks, paralleling similar efforts by competitors like OpenAI to report on unexpected AI behaviors.