Anthropic is opening a research preview of the Model Hardware Standard (MHS), a shared specification that allows AI agents to safely operate physical devices, to an initial group of scientific research labs and advanced manufacturers. MHS enables AI agents to operate multiple lab and manufacturing instruments-such as microscopes, liquid handlers, and robotic arms-in parallel, performing intricate tasks ranging from routine drug discovery experiments to laser calibration on quantum computers. The development of MHS began as a collaboration between Anthropic and HHMI Janelia Research Campus.
Typically, it takes a lab or manufacturing facility weeks or even months to set up and integrate hardware. Most devices lack the ability to communicate with each other, instead requiring specialists to build custom integrations. MHS reduces this integration work to hours or minutes. By incorporating AI into these tools, MHS also helps researchers and engineers more readily orchestrate autonomous, round-the-clock experiments and workflows, with agents able to reason through each step in an experiment, update parameters in real time, and in some cases recover from hardware errors without human intervention.
Anthropic is sharing an early version of MHS with partners across science, robotics, electronics, and manufacturing so they can collaborate on building safety evaluations and developing best practices for AI systems that operate physical equipment, ahead of making the standard open source. MHS works with any device that has a programmable interface. It is also model-agnostic, and any agent harness can access it using standard protocols, such as the Model Context Protocol. Those interested in joining the research preview can apply here.
How MHS Works
Getting multiple devices in a lab or on a factory floor to communicate with one another is challenging, even before factoring in AI integration. Each device tends to have its own programming interface, and until now there has been no standardized way to integrate them. Once the devices are connected, there is no common way for them to share data with an AI agent or to allow the agent to operate them safely.
MHS addresses these challenges by introducing a standardized driver: software that translates between a computer's operating system and a hardware device. The MHS 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 and act on. It also makes each device discoverable in a standard format, so that devices and agents can find each other and communicate across networks without requiring a custom "translator" program in between.
The MHS driver also helps an AI agent understand how to use a device it has never encountered before, providing information about machine characteristics that may not be discernible from code alone (for example, the weight of a robot arm, which matters for safe manipulation). Much of this information has historically been stored in paper manuals, on user computers, or as tacit knowledge. The MHS driver contains tags that let users write this information directly in natural language (users can do this themselves or by chatting with an agent that interviews them about their hardware setup). Using the information from these tags, the MHS driver automatically produces a reference file with information about a device's general characteristics-what it can measure, what can be adjusted, and what safety limits will be enforced. This file gives the agent everything it needs to operate the device.
After the devices are connected and the agent knows how to use each one, it needs a way to control the hardware. For MHS, there are three such mechanisms: MCP, the 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 receives operating data from each one and supervises and directs the work at a high level. The agent can sequence steps across instruments, monitor results, and adjust parameters as conditions change in real time. When the agent needs to execute long-running tasks or operate devices faster than its online reasoning would allow, it can chain together driver commands from one or more devices in code files. This allows the devices to carry out operations themselves without the agent needing to reason at every step.
During testing of MHS, Anthropic found that Claude interacts with experiments and hardware in an exploratory manner, much like a scientist would. For example, Claude was observed making an adjustment to a laser, observing the results through a camera to assess how the adjustment moved the laser beam, and repeating the process to understand the sequence of events. Claude then packaged what it learned into code files, writing a deterministic script that let it align the laser without having to reason at each step, so the entire process could run as a single command.
Early Examples from MHS
As Anthropic developed MHS, the standard was shared with a handful of labs and hardware manufacturers in biotech, robotics, quantum computing, and other fields. Across these early projects, MHS reduced device integration time, made it possible to iterate faster in various experimental settings, and assisted with live machine operation and real-time fault detection.
Genentech: Implementing MHS for Lab Automation
Researchers at Genentech implemented and tested MHS as a proof of concept for automating the BCA protein assay, a standard procedure for measuring total protein concentration in a sample. This required coordinating across a liquid handler, a robotic arm, and a plate reader. Using Claude to orchestrate the protocol, the team had the AI agent autonomously optimize fluid dynamics for both plain water and viscous protein samples. Claude independently executed trial runs, analyzed plate reader data, and converged on optimal flow rates-parameters that automation experts confirmed were reasonable for the setup. The agent also recovered from several unexpected errors on its own, though the experiments highlighted current limitations: Claude initially struggled with physical intuitions, such as recognizing that errors caused by foaming in samples were physical failures rather than software bugs.
University of Washington Baker and Pinglay Labs: Bringing AI Agents to the Bench
Zihao Song, a PhD student in the University of Washington Baker and Pinglay labs, used MHS to build a dashboard for remotely monitoring instruments, an AI agent-supervised qPCR that watches amplification curves and halts the procedure at the right moment, and an integration between a robotic arm and a liquid handler for collision-free plate handoffs. Connecting six instruments through MHS took under a week, including the time spent writing drivers. The system allowed the researcher to monitor experiments from a laptop or mobile phone and freed up time previously spent tethered to the bench.
Carnegie Mellon University: Determining Dose-Response Curves Through Rapid Automation
Researchers at Carnegie Mellon University used MHS to run serial dilution dose-response experiments about three times faster than before, with a Claude Opus 4.8 agent orchestrating a liquid handler, a plate reader, a robotic arm, and monitoring cameras spread across three computers with fundamentally incompatible interfaces. Driver development took about eight hours compared to the several weeks a vendor-built setup typically requires. The agent evaluated resulting curves, independently decided to discard a poor-fitting result, reran the experiment with adjusted parameters, and produced a strong, usable fit with no human input.
HHMI Janelia: Using MHS to Accelerate Microscopy Research
At HHMI Janelia Research Campus, researchers are using MHS to speed up a range of microscopy-related projects. Scientist Virginie Ruetten used MHS to unify and orchestrate a rig that previously involved seven different vendor programs without a shared interface. MHS replaced point-to-point connections with a single interface, reducing new hardware integration from a multi-day project to a few minutes. The standard also enabled quantitative monitoring and online analysis across multiple data streams, and facilitated agentic microscopy where Claude can identify regions of interest and adjust imaging parameters autonomously.
QuEra Computing: Using MHS in Quantum Laser Stabilization
QuEra, which builds quantum computers using neutral atoms, used MHS to give an AI agent control over parts of the laser system inside its quantum machines. The agent developed a controller that recovers the laser's "lock"-the ultra-precise frequency lasers must hold to interact with atoms-99.3% of the time without human intervention. The agent ran hundreds of iterations overnight, reducing recovery time from 5–10 minutes (manual) or 150 seconds (scripted) down to about six seconds, while improving the success rate from 58% (scripted) to 99.3%. The agent also tuned the laser's servo parameters to be roughly ten times quieter than the existing expert-set configuration.
Tetsuwan Scientific: Using MHS to Run qPCRs to Profile Local Pollution
Researchers at Tetsuwan integrated MHS with its automated biology lab platform, ResearchOS, to orchestrate a qPCR workflow contributing to citizen science efforts to characterize pollution in California's San Pedro Creek. MHS enabled camera-based detection of pipetting errors such as bubbles and foam, with Claude suggesting and executing automated recovery strategies. The integration also helped refine compiler heuristics for transfer precision, with Claude-assisted models predicting multi-dispense precision roughly 12% more accurately than manufacturer specifications.
Industry Partners Building MHS Support
Hardware vendors and software companies are also building MHS support into their equipment:
- Amazon Web Services will support MHS through Strands Robots, providing participants a private, pre-release version of the Strands Robots package for the research preview.
- Automata is adding MHS support to LINQ, their lab automation platform, for intelligent error handling.
- Danaher and Anthropic are exploring how MHS-supported capabilities could enable smart instruments and autonomous laboratories.
- Doosan Robotics is testing MHS with their robotic arms for automated quality assurance and multi-robot coordination.
- MBF Bioscience is building an MHS driver for ScanImage, software that runs laser-scanning microscopes in hundreds of neuroscience labs.
- QIAGEN is experimenting with MHS through a proof of concept on its QIAsymphony Connect nucleic acid purification platform.
- Tecan is adding MHS support for their Fluent liquid handling platforms.
- Universal Robots has had early access to MHS and plans to add support to its robotics platform.
Joining the Research Preview
These early results are encouraging, but additional work remains before the standard is open-sourced. As a large language model, Claude learns about the physical world through text and images, meaning its spatial and physical reasoning have limitations that still require expert oversight. MHS also doesn't yet work with hardware that lacks a programming interface, so Anthropic is working with manufacturers of such devices to build in MHS drivers.
Many developers already use Claude Code to work with individual pieces of physical equipment; for the next phase of MHS, Anthropic hopes to expand the standard to cover more devices. Early adopters include Hugging Face, which is adding MHS support in LeRobot, its robotics library, and Raspberry Pi, which is enabling MHS integration across a number of its products following successful tests using their Camera MHS Driver.
Anthropic will also use the research preview to build additional safety evaluations with launch partners and strengthen protections for AI use in the physical world. A physical safety roadmap is being developed to further bolster Anthropic's safeguards policy and enforcement coverage against the risk of misuse. When MHS is open-sourced, findings from the research preview will be released as part of guidance for deploying the standard safely.
Stakeholders across industries are invited to join the waitlist for the research preview of MHS. Interested parties can submit their interest here.
Acknowledgments
MHS began as a collaboration between Alek Kemeny on Anthropic's Beneficial Deployments team and Arco Bast, a postdoctoral scientist at HHMI Janelia Research Campus. Bast was running complex brain-imaging experiments on a rig combining lasers, motorized focusers, and specialized cameras from different vendors with no common interface. To speed up his experiments, he developed a shared memory dictionary enabling the instruments to communicate at memory speed. Kemeny and Bast then worked together to integrate AI models into that interface.