Previewing the Model Hardware Standard
Anthropic has launched a research preview of the Model Hardware Standard, a shared specification enabling AI agents to operate lab and manufacturing instruments—including microscopes, liquid handlers, and robotic arms—in parallel. The standard emerged from a collaboration with HHMI Janelia Research Campus and is designed to cut hardware integration timelines from weeks to hours while giving AI agents the ability to orchestrate autonomous, around-the-clock experiments.
What changed
Anthropic has opened a research preview of the Model Hardware Standard, a specification that lets AI agents operate physical laboratory and manufacturing equipment safely and in parallel. The initial cohort includes scientific research labs and advanced manufacturers working with instruments such as microscopes, liquid handlers, and robotic arms. The development originated from a partnership between Anthropic and HHMI Janelia Research Campus, where researchers previously struggled to unify devices from different vendors that lacked any common interface. MHS is model-agnostic and accessible through standard protocols including the Model Context Protocol, allowing any compatible agent harness to interface with supported hardware.
Traditional lab and factory setups typically require weeks or months of integration work, often demanding specialist knowledge to build custom connections for each device. MHS reduces this burden to hours or minutes by providing a standardized driver layer that translates between a computer's operating system and hardware devices using simple primitives like read and write commands. Each device becomes discoverable in a standard format, enabling agents and instruments to locate and communicate with one another across networks without bespoke intermediary software. Beyond connectivity, the driver captures device characteristics—such as a robot arm's weight or safety thresholds—that would otherwise exist only in paper manuals or tacit knowledge, allowing AI agents to operate unfamiliar equipment safely.
Why it matters
Once connected, agents can sequence operations across multiple instruments, monitor results in real time, and adjust parameters as conditions change. For tasks requiring speed beyond online reasoning capacity, agents can pre-chAIN driver commands into code files that execute autonomously. Early testing revealed Claude interacting with experiments in an exploratory fashion similar to a human scientist: adjusting a laser, observing the outcome through a camera, and iterating until achieving alignment, then packaging that learned behavior into a deterministic script. Partners including Genentech, Carnegie Mellon University, QuEra, and HHMI Janelia have demonstrated reductions in integration time, faster experimental iteration, and real-time fault detection across applications ranging from protein assays to quantum computer laser calibration.
Hardware vendors are also building native MHS support into their platforms. AWS plans to offer MHS through its Strands Robots library, while Automata is adding support to its LINQ lab automation platform. Danaher is exploring how MHS could scale smart instruments and autonomous biomedical labs, Doosan Robotics is testing the standard with quality-assurance tasks across multiple robotic arms, and Universal Robots plans to add MHS support to its robotics platform. MBF Bioscience is developing an MHS driver for ScanImage, software used in hundreds of neuroscience labs worldwide, and QIAGEN has demonstrated a proof-of-concept on its nucleic acid purification system showing how agents could guide operators through recovery procedures.
What to watch
Despite encouraging early results, significant limitations remain. As a language model, Claude's spatial and physical reasoning has constraints that still require expert oversight—for instance, Genentech researchers had to help Claude distinguish physical sample failures like foaming from software bugs. MHS also cannot yet support hardware lacking a programmable interface, and the team is working with manufacturers to add driver capabilities. Future plans include expanding the standard to cover more devices developers build on, with early adopters such as Hugging Face integrating MHS into its LeRobot robotics library and Raspberry Pi enabling integration across several products. The research preview will also be used to develop additional safety evaluations and a physical safety roadmap before the standard is open-sourced.