Anthropic has launched its Model Hardware Standard (MHS), a new framework designed to allow AI agents to operate physical devices in scientific research and advanced manufacturing. This marks Anthropic's first venture into "physical AI," aiming to integrate large language models like its Claude chatbot with real-world equipment such as microscopes, robotic arms, and liquid handlers.
MHS addresses a significant bottleneck in labs and manufacturing facilities where integrating hardware typically takes weeks or months due to proprietary solutions and the need for bespoke integrations. Anthropic claims MHS can reduce this integration time to hours or minutes, facilitating "autonomous, round-the-clock experiments and workflows." The standard is model-agnostic and uses a simple set of primitives like "read" and "write" commands for universal hardware understanding and communication.
Early applications and partnerships demonstrate the potential of MHS. Researchers at Genentech used MHS to automate a BCA protein assay, coordinating a liquid handler, robotic arm, and plate reader. Carnegie Mellon University researchers employed MHS to run dose-response experiments three times faster, orchestrating multiple devices across different computer systems. Quantum computing company QuEra used MHS to allow an AI agent to control parts of its laser system, achieving a 99.3% recovery rate for laser lock without human intervention. Partners like Amazon Web Services, Automata, Danaher, Doosan Robotics, MBF Bioscience, QIAGEN, Tecan, and Universal Robots are also integrating or exploring MHS support.
Anthropic is initially providing MHS as a research preview to a select group of organizations in science, robotics, and manufacturing, with plans to open-source the standard later. The company is also building a silicon team and recently hired hardware executive Caitlin Kalinowski, signaling a deeper push into hardware, similar to rivals like OpenAI and Amazon. Elizabeth Kelly, head of beneficial deployments at Anthropic, highlighted the benefits for both science and industry. Safety is a key focus, with Anthropic developing a physical safety roadmap and collaborating with partners on evaluations ahead of the open-source release.
