Summary
Anthropic published Project Fetch Phase Two, reporting that Claude Opus 4.7 completed a set of robotics tasks with an off-the-shelf quadruped without human assistance and far faster than prior human-team baselines. The update is a concrete capability signal for embodied AI, even as Anthropic notes that precision movement remains unsolved.
What changed
Anthropic published a new Project Fetch Phase Two research update showing autonomous Claude Opus 4.7 performance on physical robotics tasks.
Why it matters
The result moves AI autonomy discussion from software-only agents toward physical-world execution. For robotics, manufacturing, and lab automation teams, it suggests frontier models are becoming more useful as planning and tool-control layers, while still requiring careful safety boundaries and human oversight for precision and risk.
Evidence excerpt
Anthropic's Frontier Red Team describes Claude Opus 4.7 operating without human assistance on robotics tasks and compares performance against Phase One baselines.