According to CoinMeta, the progress of AI for Anthropic has almost doubled. Metr indicates that since October 2025, its baseline capability has increased by 1.9 times. This AI security non-profit organization found during an 11-week assessment that AI itself drove approximately 80% of the acceleration, and there is no confidence in the scalability of training computing resources. Metr also found no convincing evidence of any major bottlenecks that could slow down progress. Nevertheless, AI is still unable to make fully autonomous contributions during the most critical stages of model training, with an estimated uncertainty ranging from 1.5 to 2.8 times within a 95% confidence interval.