Anthropic on Thursday opened a research preview of the Model Hardware Standard (MHS), a shared specification designed to let AI agents safely operate physical devices such as microscopes, liquid handlers and robotic arms, according to an announcement published on the company's website.
The company is giving a first group of scientific research labs and advanced manufacturers access to the preview, and says it plans to make the standard open source once safety evaluations and best practices have been developed with partners across science, robotics, electronics and manufacturing.
Solving the integration bottleneck
Anthropic says the project began as a collaboration with HHMI Janelia Research Campus, the research campus of the Howard Hughes Medical Institute. The problem it targets is familiar to anyone who has worked in a laboratory or on a factory floor: most instruments do not communicate with each other, and integrating them typically takes weeks or months of specialist work building bespoke connections.
With MHS, Anthropic says that integration work shrinks to hours or minutes. Once devices are connected, AI agents can orchestrate autonomous, round-the-clock experiments, reasoning through each step of an experiment, updating parameters in real time and, in some cases, recovering from hardware errors without human intervention. The company says the standard has already been used for tasks ranging from routine drug discovery experiments to laser calibration on a quantum computer.
For more coverage of the tools reshaping AI development, AI Buzz Wire tracks every major release across the industry.
How the standard works
At the core of MHS is a standardized driver: software that translates between a computer's operating system and a hardware device. The driver exposes a simple set of primitives — commands such as "read" (for example, get temperature) or "write" (set temperature) — that any hardware device can understand and act on.
Each device is also made discoverable in a standard format, so agents and instruments can find each other across a network without a custom translator in between. According to Anthropic, the driver additionally gives an agent the contextual knowledge that has traditionally lived in paper manuals or in a technician's head — for example, the weight of a robot arm, which matters for manipulating it safely. Users can write that information into natural-language tags themselves, or chat with an agent that interviews them about their hardware setup. The driver then automatically produces a reference file describing what the device can measure, what can be adjusted and which safety limits will be enforced.
Agents control hardware through three mechanisms: the Model Context Protocol (MCP), a command line interface, and code files containing APIs. Together, Anthropic says, these allow an agent to orchestrate multiple devices with a single line of code, sequence steps across instruments, monitor results and adjust parameters as conditions change.
What Anthropic observed with Claude
In testing, Anthropic reports that Claude interacted with hardware in an exploratory way that resembles how a scientist works. In one example, the model adjusted a laser, observed the results through a camera to see how the beam moved, and repeated the process to understand the sequence of events. Claude then packaged what it had learned into a deterministic script, allowing the alignment procedure to run as a single command without the model reasoning through every step.
That pattern — an agent exploring interactively, then compiling its findings into repeatable code — is central to how Anthropic expects MHS to be used for long-running tasks that would be slow or expensive to supervise continuously.
Model-agnostic by design
Anthropic is positioning MHS as infrastructure rather than a product lock-in. The standard works with any device that has a programmable interface, and the company says it is model-agnostic: any agent harness can access it using standard protocols such as MCP. That could matter for adoption in research labs, where equipment from multiple vendors and models from multiple providers are often mixed.
The announcement lands amid intensifying competition to connect AI models to the physical world, with rivals pursuing robotics-focused models and hardware partners. By targeting laboratories and manufacturing lines first — environments with structured, instrument-heavy workflows — Anthropic is betting that scientific and industrial automation will be the first commercially meaningful domain for embodied agents.
What to watch
The research preview is limited to an initial group of partners, and Anthropic has not announced a date for the open-source release. Key questions for the coming months include how many device manufacturers build MHS drivers, what the safety evaluations developed with partners look like, and whether competitors adopt or interoperate with the standard.
For laboratories weighing adoption, the pitch is straightforward: hardware integration that once took months could become a configuration task measured in hours, with agents able to run experiments continuously in the background. Whether that promise survives contact with messy, real-world equipment is the test the research preview is designed to answer.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
