Anthropic proposes plumbing spec to link AI agents to lab kit and robots

Anthropic, the AI safety startup, has proposed a new protocol called the Model Hardware Standard (MHS) to enable AI agents to safely control physical devices in laboratories, factories, and robots. This optimistic ambition is challenging, as AI models currently lack reliable predictive capabilities. The MHS protocol is similar to the Model Context Protocol (MCP), which allows AI models to connect to data sources.

Anthropic envisions MHS allowing AI agents like Claude to operate equipment in various settings, from laboratories to factories. The protocol is currently being tested at the Howard Hughes Medical Institute's Janelia Research Campus in Maryland. Other entities with relevant laboratory and industrial equipment can apply to participate in the research preview.

Setting up and integrating hardware in a lab or manufacturing facility typically takes weeks or months. MHS aims to reduce this integration work to just hours or minutes. To connect hardware to an AI model, a programmable interface is required. Many industrial machines already offer such an interface, but the ecosystem of these devices is diverse, making it difficult for developers to create custom hookups.

MHS serves as a universal translation layer, using a limited set of primitives like "read" and "write," similar to how simple tools like Bash can be used to power AI agents. The driver software makes connected devices discoverable in a standard format and supports tags that convey information about device functions. Users can provide this data by conversing with the model during setup. The tags enable the driver to produce a reference file detailing device characteristics.

AI agents can interact with devices using three control paths: MCP, command line interface, and API code. They can carry out commands on connected instruments, monitor test results, or adjust knobs and dials. Companies like Genentech and QuEra have already used MHS to enhance their operations, with Genentech running a drug-discovery experiment and QuEra improving laser stabilization.

Major tech companies, including AWS, Automata, Danaher, Doosan Robotics, MBF Bioscience, Qiagen, Tecan, and Universal Robotics, are planning to support MHS in their respective products. Anthropic acknowledges that there is still more work to do before open-sourcing MHS, as large language models (LLMs) still lack physical intuition and must be built with safety evaluations and protections in mind. The research preview aims to strengthen these protections and build more safety evaluations for using AI in the physical world.