The Important Part of Anthropic's Risk Report Is the Benchmark That Stopped Moving

Anthropic Raises AI Risk Concerns as Claude Models Show Early Signs of R&D Acceleration

Claude writes most production code at Anthropic, accelerating internal R&D, but misalignment risk rose from "very low" to "low" as model refusals and data leakage emerged. Growing velocity-safety gap signals governance and control challenges in autonomous AI-driven development.