Introducing the model hardware standard for ai-driven physical operations

Anthropic and HHMI Janelia Research Campus have launched a research preview of the Model Hardware Standard (MHS), a shared specification designed to enable AI agents to safely operate physical devices. The initiative aims to streamline the integration of lab and manufacturing instruments such as microscopes, liquid handlers, and robotic arms, allowing for parallel operations and complex tasks like drug discovery and quantum computer calibration. MHS reduces integration time from weeks to minutes by providing a standardized interface that allows AI agents to interact with devices through simple commands like 'read' and 'write'. The MHS driver, a key component of the standard, translates between a computer's operating system and hardware devices, making each device discoverable in a standard format. This enables devices and agents to communicate across networks without the need for bespoke translation programs. The driver also provides information about device characteristics, such as weight and safety limits, which can be written in natural language or through an AI agent's interview with the user. This information is used to generate a reference file that gives the agent everything needed to operate the device. MHS supports three mechanisms for controlling hardware: the Model Context Protocol (MCP), a command line interface, and code files (APIs). These work together to enable orchestration across multiple devices via a single line of code. Once the agent can control the devices, it can receive operating data, supervise workflows, and adjust parameters in real time. For long-running tasks, the agent can chain driver commands in code files, allowing devices to operate independently. Early adopters have reported significant improvements in efficiency and safety. Researchers at Genentech used MHS to automate the BCA protein assay, while Zihao Song at the University of Washington developed a dashboard for remote monitoring and an AI-supervised qPCR process. Carnegie Mellon University researchers ran serial dilution experiments three times faster using MHS. At HHMI Janelia, Virginie Ruetten unified a microscopy rig that previously involved seven different vendor programs. QuEra, a quantum computing company, used MHS to give an AI agent control over its laser system, achieving a 99.3% recovery rate for laser lock without human intervention. Hardware vendors and software companies are also integrating MHS support into their equipment. For example, Hugging Face is adding MHS support in LeRobot, and Raspberry Pi is enabling MHS integration across several products. Despite these successes, challenges remain, including the need for MHS to work with hardware that lacks a programming interface. Developers are working with manufacturers to build MHS drivers for such devices. The initiative is part of a broader effort to enhance safety and oversight for AI in physical environments. Anthropic is developing a physical safety roadmap to strengthen safeguards against misuse. The research preview is open to stakeholders across industries, with a waitlist available for participation. Additionally, Anthropic is launching a $5 million grant program to fund independent research on AI's impact on user wellbeing. The company also shared insights into its watermarking method, addressing concerns about its effects on Claude's outputs. The MHS project began as a collaboration between Alek Kemeny from Anthropic's Beneficial Deployments team and Arco Bast, a postdoctoral scientist at HHMI Janelia Research Campus. Bast developed a shared memory dictionary to enable communication between instruments in his brain-imaging experiments. Kemeny and Bast integrated AI models into this interface, laying the foundation for MHS. The initiative has received contributions from numerous researchers and developers, including Aaron Boswell, Ben Arthur, Boaz Mohar, Gagan Bhat, Mark Kittisopikul, Nadine Yasser, Nick Purcell, Takashi Kawase, and Virginie Ruetten. Anthropic looks forward to advancing MHS with industry partners and the open-source community.