Anthropic has released a research preview of its Model Hardware Standard, a new specification that allows AI agents to operate physical laboratory and manufacturing equipment without requiring custom software for each device. The system can reduce hardware integration from weeks or months to hours or minutes, while giving AI agents a standardized way to discover equipment, understand safety limits, and coordinate multiple machines working together.

What Problem Does the Model Hardware Standard Solve?

Laboratory and manufacturing equipment typically comes with different programming interfaces, software formats, and data structures. Connecting these devices together has traditionally required specialists to build custom software for each setup. Adding an AI agent into the mix created yet another layer of integration work. The Model Hardware Standard introduces a unified driver that translates between software and physical devices, using basic commands like reading a temperature or changing a setting while also describing each machine's characteristics and safety limits.

This approach matters because laboratory conversations are tightly coupled sessions that require considerable shared context. When AI agents can access hardware through standardized mechanisms, including the Model Context Protocol, command-line tools, and APIs, they can coordinate multiple instruments, sequence their work, monitor outcomes, and adjust parameters as conditions change.

Where Is the Hardware Standard Already Being Tested?

Anthropic has moved beyond simulations into real-world testing across scientific research and advanced manufacturing. Early projects span drug discovery, brain imaging, and quantum computing. At Genentech, AI agents using the standard ran a drug-discovery experiment with real-time error handling. At HHMI Janelia Research Campus, researchers compressed an imaging experiment from weeks to a single day. On QuEra quantum computers, the system improved laser stabilization from 58% to 99.3%.

The Genentech test revealed both the promise and the limitations of current AI systems. Researchers asked Claude, Anthropic's AI assistant, to run a protein assay that required a liquid handler, robotic arm, and microplate reader to work together. Claude initially selected generic liquid-handling settings for both water and a more viscous protein solution, which caused bubbles to form and produced inaccurate transfers. When researchers asked Claude to optimize the flow rate itself, the system tested different settings, analyzed readings, and eventually selected a flow rate of around 140 microliters per second for water and 10 microliters per second for the protein sample, which Genentech's automation specialists judged reasonable.

What Are the Key Capabilities of the Hardware Standard?

• Standardized Driver Layer: Translates between software and physical devices using basic commands while describing machine characteristics and safety limits that might not be obvious from software alone, such as the weight of a robotic arm or safe operating ranges for parameters.
• Multi-Device Coordination: Allows AI agents to receive data from several instruments simultaneously, sequence their work, monitor outcomes, and change parameters as conditions change without requiring separate integrations for each device.
• Model-Agnostic Design: Works with any device that has a programmable interface, meaning the standard is not locked to specific equipment manufacturers or AI models.
• Rapid Integration: Reduces hardware integration time from weeks or months to hours or minutes, enabling faster deployment of AI-driven automation in research and manufacturing settings.

What Did Researchers Learn About AI's Physical Limitations?

The Genentech experiment provided a useful reminder of what current AI systems still struggle with. Claude recovered independently from some equipment errors, including failures involving pipette tips and fluid detection. But when bubbles caused problems during mixing, its initial response was to retry the operation with different parameters in the same well, which made the problem worse. Researchers had to explain that the error came from the physical behavior of the liquid rather than a software fault, and tell the system to move to a clean well and reduce mixing.

Anthropic acknowledges that large language models learn about the physical world mainly through text and images and can lack the spatial and physical intuition required for some real-world tasks. This is why the company is keeping the Model Hardware Standard in a research preview before making it openly available, using failures to justify additional safety evaluations.

How Are Industry Partners Contributing to the Standard?

The Model Hardware Standard began through a collaboration between Anthropic's Beneficial Deployments team and HHMI Janelia Research Campus, where researcher Arco Bast was working with a brain-imaging setup combining lasers, cameras, and motorized equipment from different manufacturers. That collaboration has since expanded into robotics, electronics, biotechnology, and manufacturing. Hardware companies including Automata, Doosan Robotics, MBF Bioscience, QIAGEN, Tecan, and Universal Robots are among those testing or building support for the standard. AWS is supporting the preview through Strands Robots, Hugging Face is adding MHS support to LeRobot, and Raspberry Pi is enabling integration across a number of its products following tests with a Camera MHS Driver.

At the University of Washington, PhD student Zihao Song used the Model Hardware Standard in the Baker and Pinglay labs to connect equipment used in protein research. His experiments included a dashboard that allowed instruments to be monitored remotely, an AI-supervised quantitative PCR workflow, and coordination between a robotic arm and liquid handler. In the qPCR test, the system monitored amplification curves as they were produced and could halt the reaction when instructed before moving the equipment into a holding stage. Song also reported connecting six instruments through the standard in under a week, including the time required to write drivers for them.

Steps to Implement Hardware Integration With AI Agents

• Assess Your Equipment: Identify which laboratory or manufacturing devices have programmable interfaces and would benefit from AI coordination, starting with equipment that currently requires manual handoffs or custom software integration.
• Join the Research Preview: Apply to participate in Anthropic's research preview program if you are a researcher, manufacturer, or industry stakeholder interested in testing the Model Hardware Standard with your equipment.
• Plan for Safety Evaluation: Prepare to work with Anthropic on physical safety evaluations and testing, as the company is gathering evidence on how AI behaves when given direct control over physical equipment before full open-source release.
• Document Equipment Specifications: Compile detailed information about your devices' operating limits, safety constraints, and physical characteristics so they can be properly represented in the standardized driver layer.

The Model Hardware Standard is not yet open source. Anthropic is first making it available to a group of scientific research labs and manufacturers so it can expand safety evaluations and gather evidence on how AI behaves when given direct control over physical equipment. The company says it will use the research preview to develop additional physical safety evaluations and publish guidance when the standard is eventually released as open source.

These early demonstrations remain proofs of concept. Researchers note that more complicated laboratory protocols will require further optimization and that running AI agents continuously over long experiments also introduces compute costs. That is part of the reason Anthropic is stopping short of presenting the Model Hardware Standard as a finished automation layer. The company is currently inviting researchers, manufacturers, and other industry stakeholders to join the preview and help shape the standard's development.