What Changed: A Standard for AI-Controlled Hardware
Integrating lab or factory equipment typically takes weeks or months. Each instrument has its own programming interface, requiring bespoke integration code, and there’s no common way for devices to share data with an AI agent. On August 27, 2026, Anthropic announced a research preview of the Model Hardware Standard (MHS), a shared specification that lets AI agents safely operate physical devices like microscopes, liquid handlers, and robotic arms. The goal is to cut integration time from weeks to hours or minutes.
MHS began as a collaboration between Anthropic and HHMI Janelia Research Campus. It’s now open to a first group of scientific research labs and advanced manufacturers. The standard works with any device that has a programmable interface, is model-agnostic, and can be accessed via standard protocols like the Model Context Protocol (MCP).
How MHS Works: Standardized Drivers and Natural Language Tags
The core of MHS is a standardized driver: software that translates between a computer’s operating system and a hardware device. The driver uses a simple set of primitives—commands like “read” (e.g., “get temperature”) or “write” (e.g., “set temperature”)—that any hardware device can understand. It also makes each device discoverable in a standard format, so devices and agents can find each other across networks without a bespoke translator.
Crucially, the driver helps an AI agent understand a device it has never seen before. It contains tags that let users write machine characteristics in natural language—for example, the weight of a robot arm, which is important for safe manipulation. Users can do this themselves or by chatting with an agent that interviews them about their hardware. From these tags, the driver automatically produces a reference file with general characteristics: what the device can measure, what can be adjusted, and what safety limits are enforced. This file gives the agent everything it needs to operate the device.
Once devices are connected, there are three control mechanisms: MCP, command line interface, and code files (APIs). These work together to enable orchestration across multiple devices via a single line of code. For long-running tasks or operations faster than online reasoning allows, the agent can chain driver commands into code files, letting devices execute autonomously without step-by-step reasoning.
Early Partner Results: From Protein Assays to Quantum Lasers
Anthropic shared MHS with partners in biotech, robotics, quantum computing, and other fields. Here are some early projects:
- Genentech implemented MHS as a proof-of-concept for automating the BCA protein assay, which coordinates a liquid handler, a robotic arm, and a plate reader. Claude orchestrated the protocol and autonomously optimized liquid handling parameters. For water, it found an optimal flow rate of ~140 µL/s (0.016 RMSE); for viscous BSA, ~10 µL/s (0.181 RMSE)—parameters confirmed reasonable by automation experts. Claude also recovered from errors like tip pickup failures on its own.
- University of Washington Baker and Pinglay labs PhD student Zihao Song used MHS to build a remote monitoring dashboard, an AI agent-supervised qPCR that watches amplification curves and halts at the right moment, and a collision-free integration between a robotic arm and a liquid handler.
- Carnegie Mellon University researchers used MHS to run serial dilution dose-response experiments about three times faster than before, coordinating a liquid handler, plate reader, robotic arm, and monitoring cameras across three incompatible computers.
- HHMI Janelia scientist Virginie Ruetten used MHS to unify and coordinate a microscope setup that previously required seven different vendor programs.
- QuEra Computing let an AI agent control laser systems inside a quantum computer. The agent-developed controller recovered laser lock automatically in 99.3% of cases without human intervention.
- Tetsuwan Scientific integrated MHS into its automated biology platform ResearchOS for qPCR workflows, supporting citizen science pollution analysis in San Pedro Creek, California.
Hardware vendors are also joining: AWS will offer MHS support via Strands Robots, Automata is adding MHS to its LINQ platform, Danaher is exploring smart instruments with Anthropic, Doosan Robotics is testing robotic arm coordination, MBF Bioscience is building a driver for ScanImage, QIAGEN is running a proof-of-concept on QIAsymphony Connect, and Tecan and Universal Robots plan support.
Limitations and Trade-offs: Physical Reasoning Gaps
MHS has clear limitations. Claude, as a large language model, understands the physical world through text and images, so its spatial and physical reasoning is limited and requires expert supervision. For example, Genentech researchers had to guide Claude to recognize that bubbles in samples were a physical problem, not a software error. Claude’s default instinct was to retry with different parameters, which only agitated the fluid further. Once informed about the physics, it adapted and maintained that context.
MHS also doesn’t support hardware without a programmable interface. Anthropic is working with manufacturers of such devices to build in drivers. During the research preview, Anthropic will collaborate with partners to build safety evaluations and develop a physical safety roadmap to guard against misuse. MHS is planned to be open-sourced in the future, with findings from the research preview released as guidance for safe deployment.
Practical Takeaways for Builders
For product builders and AI tool learners, MHS offers a concrete path to integrate AI with physical devices. The key insight is that standardizing drivers and using natural language tags can dramatically reduce integration time. Instead of writing custom code for each device, you can describe device characteristics in plain language, and the driver generates a reference file for the agent.
When building with MHS, consider the following:
- Start with a programmable device: MHS works with any device that has a programmable interface. If your hardware lacks one, you’ll need to work with manufacturers or add a driver.
- Use natural language tags: Enrich your device drivers with tags describing physical characteristics. This helps the agent understand safe operation.
- Leverage code files for long tasks: For operations that require speed or long durations, chain driver commands into code files so devices can run autonomously.
- Plan for physical reasoning gaps: Don’t expect the agent to handle all physical nuances. Build in expert oversight and feedback loops to correct errors.
Anthropic is accepting applications for the research preview at modelhardwarestandard.com. If you’re building tools that interact with lab or factory equipment, this is a promising standard to watch.
The Bottom Line
MHS is an early but promising step toward making AI agents practical for physical-world automation. It reduces integration time, enables parallel operation of multiple devices, and allows agents to learn and adapt. However, it’s not a silver bullet: physical reasoning remains a challenge, and expert supervision is still necessary. As the standard evolves and becomes open source, it could unlock new levels of autonomy in labs and factories, but builders should approach it with realistic expectations about its current limits.
Sources
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.