Anthropic previews a standard letting AI agents operate lab equipment

Anthropic opened a waitlist-gated research preview of the Model Hardware Standard, a shared specification meant to let AI agents operate physical lab and manufacturing equipment — microscopes, robotic arms, liquid handlers, plate readers — without a bespoke integration built for each device. The standard defines simple “read” and “write” commands that any compliant device can expose, plus natural-language tags describing what a device does and its safety limits, which Anthropic said cuts integration time from weeks to minutes.

Anthropic likened the standard to a USB-C cable — one common interface replacing whatever proprietary connection a device would otherwise need — the same analogy it has used for its 2024 Model Context Protocol, the equivalent standard it built for connecting AI models to software tools. The hardware standard is designed to work alongside that protocol and with AI models beyond Claude, rather than exclusively with Anthropic’s own systems.

Early testing partners include Genentech and the Howard Hughes Medical Institute’s Janelia Research Campus, which had worked with Anthropic during the standard’s development. Anthropic said it plans to open-source the specification once it has built out safety evaluations and best practices with this first group of partners, rather than releasing it openly from the outset.

The preview extended Anthropic’s move into applying Claude beyond software, following its report, nine days earlier, that Claude had helped design experimentally validated protein binders — though a standard for issuing write commands to physical equipment raises a different order of safety question than a chatbot answering questions: a mistake here can move a robotic arm or heat a live sample rather than just produce a wrong sentence, part of why access started with a small, vetted group rather than an open release.