Anthropic's Claude Takes First Steps into Physical AI with Model Hardware Standard
A new framework for integrating advanced language models like Anthropic’s Claude with physical objects has been unveiled by the company. The Model Hardware Standard (MHS) is designed to connect large language models with equipment in manufacturing, science labs, and other settings. This marks a significant move into so-called physical AI for Anthropic, which typically focuses on developing and applying its language models in software environments.
With MHS, companies can integrate AI into their equipment quickly – often within hours or minutes – whereas this process usually takes weeks or even months with custom builds required by specialists. The standard is expected to pave the way for autonomous experiments and workflows that run around the clock without human intervention. Scientific research and advanced manufacturing applications are among its primary uses.
The MHS framework can connect multiple devices, enabling them to communicate through a set of commands like ‘read.’ Any hardware device can understand these commands and act on them. This standard is model-agnostic, meaning it works with any large language model – not just Claude – including models built by other companies such as OpenAI or open-source models.
MHS is built upon the Model Context Protocol (MCP), a universal, open standard for connecting data sources that Anthropic debuted in 2024. The MCP serves as a kind of ‘USB’ for AI to software connection, facilitating seamless interaction between different systems and devices. According to Alek Kemeny, a member of the technical staff at Anthropic, not all existing equipment can connect directly to MHS due to lack of programming interfaces.
As part of this project, Anthropic is collaborating with numerous device manufacturers to develop new products that come pre-loaded with the necessary interface for connecting to MHS. The company is also assisting in adding these connections to existing products. This effort aims to streamline the process of integrating AI into various settings and applications.
Anthropic’s partnership with HHMI Janelia Research Campus, a biomedical research center in Virginia, played a crucial role in developing MHS. A small group of labs and hardware manufacturers received early access during development, focusing on fields such as biotech, robotics, and quantum computing. Some notable partners include Genentech, Carnegie Mellon university, QuEra, Universal Robots, Amazon Web Services, Doosan Robotics, Danaher, and Hugging Face.
The MHS research preview comes at a time when interest in combining AI with robotics is on the rise. Other companies like Nvidia have also been championing physical AI, predicting that industrial companies will soon become robotics companies. Anthropic’s move into this space marks an important step forward for its Claude language model and highlights the potential of integrating advanced AI tools with various industries.
According to Jonah Cool, head of partnerships and deployment of science at Anthropic, proprietary solutions often hinder scientific progress by being brittle and inflexible. MHS offers a standardized interface that aims to connect any model to equipment easily, avoiding vendor lock-in for scientists. This development has significant implications for the future of AI integration in various settings.