Anthropic kicked off phase one of the Model Hardware Standard (MHS) research preview: a standard so AI agents can safely operate scientific and advanced-manufacturing equipment. Bespoke hardware integration that took days or weeks is cut to hours or minutes, with discoverable devices and safer control. Early tests: real-time error handling on a Genentech drug-discovery run, compressing an HHMI Janelia imaging experiment from weeks to a day, and lifting QuEra laser stabilization from 58% to 99.3%. Developers already using Claude Code on boards and cameras are the next devices MHS aims to cover.
Key Takeaways
- βCoding agents are moving from repos onto lab benches and factory floors; a hardware interface standard is the next substrate.
- βClaude Code is already driving boards and cameras; MHS is the path to one safe interface for those devices.
- βThe preview is for safety evals and physical-intuition gaps before open-sourcing with industry partners.