😺 Anthropic wants Claude operating real lab gear

Anthropic and HHMI Janelia have opened a research preview for something called the Model Hardware Standard, or MHS. The pitch is simple: give programmable lab and factory gear a shared interface so an AI agent doesn’t need custom code for every microscope, robot arm, or plate reader it meets.

Each device gets a standard driver with read and write commands, plus plain-language tags that describe what the machine does and where its safety limits are. Early partners include Genentech, Carnegie Mellon, and QuEra. At QuEra, an agent-built script recovered a quantum laser’s lock in 695 out of 700 trials.

That sounds dry until you remember how much lab automation usually depends on glue code. Today, every device tends to want its own integration. Anthropic says MHS is meant to cut that mess down by giving agents one place to discover equipment, chain tasks across devices, and turn a working procedure into a script.

The company says work that can take weeks or months to wire up might shrink to hours or minutes. That would make round-the-clock automated experiments much easier to build. But this is still a research preview, not a robot scientist with opinions and tenure.

Anthropic is also clear about the limits. Claude still needs expert oversight for physical reasoning, and MHS only works where the hardware already has a programmable interface. If the standard catches on, the big story may not be one model beating another — it may be the layer that finally lets models touch the physical world without a custom integration for everything.