Alignment ForumArticleClaude Watch
METR: Independent Assessment of AI R&D Acceleration at Anthropic in October 2026
This is the most rigorous external measurement of AI acceleration we have. Doubled improvement velocity on well-specified tasks means Anthropic's training and fine-tuning pipeline is working. For builders using Claude: expect faster capability gains and more frequent breaking changes. For investors: this validates the bet on applied capability growth over raw scale. The risk section matters too. Read the full report.