Anthropic is testing a new software standard designed to allow artificial intelligence assistants such as Claude to interact more effectively with robots, scientific instruments and manufacturing hardware, marking a significant step in the company’s effort to move AI beyond screens and into the physical world.
The AI startup announced its Model Hardware Standard, or MHS, as a research preview on Thursday, giving developers in scientific, robotics and manufacturing fields an opportunity to test how Claude interacts with physical equipment and establish safeguards before the framework is released more broadly.
At its core, MHS is intended to give AI systems the information they need to understand how hardware operates without relying on paper manuals or specialised knowledge held by only a small number of experts. A manufacturer of factory robot arms, for example, could specify through MHS how an AI system should safely operate a heavy arm, including restrictions on its movement speed or operating angles.
The approach comes as businesses and investors show growing interest in applying advances in AI to the physical world. Manufacturing, robotics and scientific experimentation are emerging as major areas of interest, with a Barclays report from earlier this year forecasting that AI-powered robots and autonomous machines could develop into a trillion-dollar market by 2035.
Anthropic is entering a field in which other major technology companies are also developing AI systems and software for robots. Alphabet’s Google, OpenAI and Nvidia are among those pursuing technologies aimed at connecting artificial intelligence with physical machines, increasing competition over how AI can be deployed outside traditional computing environments.
Industry stakeholders can join a waitlist to trial MHS, Anthropic said. The company intends to open-source the framework after the research preview, allowing hardware manufacturers to develop their own MHS specifications. Anthropic has not provided a specific timeline for the wider release.
For scientific research, the company sees the technology as a way of reducing the technical barriers that can prevent researchers from using sophisticated equipment. Jonah Cool, a head of partnerships and deployment at Anthropic’s life sciences arm, said that in some cases scientific work does not happen because researchers cannot use equipment or find it too technically challenging. MHS, he said, could allow Claude to act as an enabler by helping scientists use appropriate equipment in an expert manner.
The initiative builds on Anthropic’s earlier work to establish common standards for AI interaction with software. The company introduced the Model Context Protocol, or MCP, to allow AI assistants to interact more easily with third-party applications. The standard has enabled Claude and rival chatbot ChatGPT to connect with services including Gmail, Google Calendar and Slack.
Anthropic now intends to extend that principle to physical equipment. Alek Kemeny, a member of technical staff at Anthropic, described MHS as what MCP did for software, but for the hardware world. He said its value would be particularly apparent in scientific laboratories, where researchers may work with dozens or even hundreds of different devices.
Anthropic has already been testing the framework with industry stakeholders including Amazon Web Services, Danaher, Hugging Face and Raspberry Pi. The company has also presented a demonstration intended to show how the system could function in practice.
In the promotional video, Anthropic said a scientist from biotech firm Genentech provided Claude with a PDF document describing an experiment. According to the company, Claude was then able to autonomously execute the experiment using hardware equipped with the MHS specification.
That demonstration points to the broader ambition behind the framework: reducing the distance between an AI assistant that can understand instructions and the machines capable of carrying them out. If the system can be made reliable and safe, MHS could give AI models a more direct role in laboratories, factories and robotic environments, while allowing hardware manufacturers to describe the limits within which those systems can operate.
For Anthropic, the test is therefore not simply about teaching Claude to understand machines. It is about creating a common framework through which AI can interact with increasingly complex physical systems without depending entirely on specialist human knowledge.
