.png){kind=logo}
Anthropic introduced the Model Hardware Standard (MHS) on August 27, 2026, opening a research preview for selected laboratories and manufacturers. The framework lets AI agents control physical equipment, including microscopes, robotic arms and machines that handle liquids.
The project began through a collaboration with HHMI Janelia Research Campus. Researchers there needed lasers, cameras and motorised focusing equipment to communicate.
Anthropic says laboratories and factories often spend weeks or months connecting equipment through custom software. MHS aims to shorten that work to hours or minutes by giving devices a shared way to communicate. Devices and agents can discover each other across networks without custom translators.
The system uses drivers, which translate commands between computers and machines. These commands allow agents to read measurements or change settings. Users can also describe equipment characteristics and safety limits in plain language.
The driver then creates a reference file listing device measurements, adjustable settings and operating limits. This gives agents information that might otherwise sit in manuals or depend on staff knowledge.
“MHS works with any device that has a programmable interface,” Anthropic said. The standard supports different AI models, allowing developers to use agents beyond Claude.
Agents access connected machines through the Model Context Protocol, command-line tools or programming interfaces. They can coordinate equipment, track results and adjust settings during experiments.
Amazon Web Services plans to support MHS through Strands Robots, its software library for connecting AI agents to physical devices. Participants will receive a private version of that package during testing.
Meanwhile, Doosan Robotics is testing robotic arms for automated quality checks and tasks involving several robots. Universal Robots also plans to add support to its platform.
QIAGEN has developed a working test on QIAsymphony Connect, its system for purifying genetic material. The project explores how agents could identify equipment problems and guide operators through recovery.
Tecan is adding support to its Fluent liquid handling platforms. Additionally, Automata is integrating MHS into its LINQ platform to help automated laboratories handle instrument errors.
These projects cover different stages of development, from planned support to working demonstrations. Danaher and Anthropic are also exploring how connected instruments and automated laboratories could support biomedical research and development.
ALSO READ: Top Robotics Trends to Watch in 2027
Anthropic also described a test where Claude adjusted a laser and used camera images to check how the beam moved. After repeating the process, Claude wrote a script that could reproduce the alignment through a single command.
Such scripts let machines run sequences without waiting for the agent to reason through every step. This supports longer tasks and operations that need faster execution.
However, physical reasoning still presents limits. Genentech researchers had to explain that foam in protein samples caused physical failures requiring changes beyond software fixes.
Equipment without a programming interface cannot currently use MHS. Anthropic is working with manufacturers to extend support, while Hugging Face and Raspberry Pi are developing integrations.
The company will use the preview to build safety tests with partners and strengthen protections against misuse. It also plans to publish preview findings and deployment guidance when it releases MHS as open source. Research organisations and equipment makers can apply to join the preview through the project’s website.
