Anthropic's Model Hardware Standard Lets AI Agents Autonomously Run Lab Experiments Across Drug Discovery
核心洞察
Anthropic (搜索) opened a research preview of its Model Hardware Standard (MHS), the company's first tool designed to operate in the physical world, enabling AI agents like Claude (搜索) to control scientific equipment and automate experiments.
Genentech used MHS and Claude (搜索) to autonomously optimize fluid dynamics for a BCA protein assay, with Claude independently determining optimal flow rates of ~140 µL/s for water and 10 µL/s for BSA.
Carnegie Mellon University researchers ran serial dilution dose-response experiments roughly three times faster using MHS, achieving a usable fit (R² > 0.98) after an autonomous rerun with no human input.
Anthropic (搜索) has opened a research preview of its Model Hardware Standard (MHS), a framework that allows AI agents such as Claude (搜索) to control scientific equipment, automate experiments, and manage workflows across laboratories. The technology is Anthropic's first tool designed to operate in the physical world, enabling AI agents to autonomously operate devices ranging from complex microscopes to liquid handlers, lasers, and robotic arms.
The development of MHS began as a collaboration between Anthropic (搜索) and HHMI Janelia Research Campus (搜索). Since then, Anthropic has been testing the technology with a small group of labs and hardware manufacturers in biotech, robotics, and quantum computing, including the Howard Hughes Medical Institute, Carnegie Mellon University, Genentech, and QuEra (搜索). The research preview comes as the company seeks to expand new applications into manufacturing, robotics, and pharmaceuticals, ahead of a planned IPO that could value the company at $2 trillion.
A Standard Interface Between AI and Laboratory Hardware
Anthropic (搜索) explains that a standardised driver is 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" (for example, "get temperature") or "write" (for example, "set temperature") — that any hardware device can understand and act on. It makes each device discoverable in a standard format, so that devices and agents can find each other and communicate across networks without needing a bespoke "translator" program in between.
MHS can be controlled by three 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 is able to receive operating data from each one and supervise and direct the work at a high level.
Genentech Automates the BCA Protein Assay
Researchers at Genentech implemented and tested MHS as a proof of concept for automating the bicinchoninic acid (BCA) protein assay, a standard procedure used to measure total protein concentration in a sample. The procedure involves coordinating across a liquid handler, a robotic arm, and a plate reader, with Claude (搜索) acting as a central communication hub for the hardware. All experiments were conducted in standard 96-well microplates.
The BCA assay involves handling liquids with different physical properties, ranging from simple aqueous reagents to viscous, foamy protein samples. Genentech used bovine serum albumin (BSA) at known concentrations as a reliable standard. Because these fluids behave differently under pressure and flow, pipetting must be done extremely precisely to ensure an exact volume of solution is transferred.
As a starting point, the team gave Claude (搜索) the standard BCA assay protocol to establish a baseline. In this first test, Claude selected generic liquid handling parameters with the same flow rate for both aqueous and viscous solutions, which caused bubbles to form in the viscous solution and resulted in inaccurate liquid transfers. The researchers then asked Claude to autonomously optimize fluid dynamics for both plain water and viscous protein samples, prompting the model with an experimental design to optimize the liquid transfer flow rate by conducting trial transfers with dyed liquid and taking absorbance readings with the microplate reader.
Claude (搜索) independently executed these trial runs and analyzed the resulting plate reader data to get closer to expert-performed transfers. For water, Claude concluded that a flow rate of ~140 µL/s was optimal (0.016 RMSE); for BSA, it arrived at 10 µL/s (0.181 RMSE) — parameters that Genentech's automation experts confirmed were reasonable for their setup. Ordinarily, performing this optimization requires an automation specialist to write custom programming logic for every single parameter set.
During the experiment, Claude (搜索) encountered several unexpected errors, including tip pickup failures and fluid detection errors, but managed to recover on its own — a capability that current scientific instruments mostly lack. However, the experiments also highlighted the current limits of AI models. Although they excel at general-purpose reasoning, they still struggle with physical, chemical, and biological constraints, particularly when troubleshooting errors that call for real-world physical intuition.
Carnegie Mellon Accelerates Dose-Response Experiments
Researchers at Carnegie Mellon University used MHS to run serial dilution dose-response experiments about three times faster than before, with an AI agent orchestrating a liquid handler, a plate reader, a robotic arm, and monitoring cameras spread across three computers with fundamentally incompatible interfaces.
Serial dilution is a key component of drug development used to determine dosage. The process is time-consuming and error-prone, typically requiring multiple iterations to determine the right maximum concentration and the appropriate step size between dilutions. When done by hand, setting up and conducting a set of serial dilution experiments can take weeks.
The Carnegie Mellon system combines a CyBio Felix liquid handler, a Varioskan LUX plate reader, a robotic arm to move 96-well plates, and monitoring cameras with an AI-controlled orchestrator to automatically and dynamically measure dose-response curves. Using MHS, the team developed drivers from scratch for each instrument and an orchestration layer that lets a Claude (搜索) Opus 4.8 agent run the full protocol autonomously. This took about eight hours, versus the several weeks a vendor-built setup typically takes.
To verify that MHS would operate safely and correct itself like a human operator would, the researchers artificially induced six different conditions: missing plate, rotated plate, reader busy, disconnected camera, unreachable device, and active emergency stop. The system correctly blocked all six before any device moved. The team then asked the agent to run the serial dilution experiment to achieve an acceptable curve. The model evaluated the resulting curve but found a fit too poor to accept (R² < 0.9, driven by saturation in the upper concentration range) and decided independently to discard the plate and rerun on a fresh plate with a compressed concentration range (200 µg/mL top concentration reduced to 100 µg/mL). The second run produced a strong, usable fit (R² > 0.98 with 3.4 variation across repeated measurements) with no human input at any point.
Broader Applications and Ecosystem Support
At HHMI Janelia Research Campus (搜索), researchers are using MHS to speed up a range of microscopy-related projects. Virginie Ruetten, a scientist in the Ahrens lab who studies how sleep helps the body recover from stress, used MHS to unify and orchestrate a rig that previously involved seven different vendor programs without a shared interface. Her work uses young zebrafish and an approach called WHOLISTIC imaging to capture cellular activity throughout the brain and body of a living animal using two-photon microscopy.
QuEra (搜索), a company that 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 the lasers must hold to interact with the atoms — 99.3% of the time without human intervention. Across 700 trials, the finished script recovered the correct lock 695 times.
Amazon Web Services will support MHS through Strands Robots, the library for connecting AI agents to physical devices, while Automata is adding MHS support to LINQ, their lab automation platform, to perform intelligent error handling of instruments in autonomous labs. With Danaher (搜索), Anthropic (搜索) will actively explore how MHS-supported capabilities could enable its smart instruments and autonomous laboratories to scale biomedical research and development.
Limitations and Next Steps
MHS does not yet operate with hardware that lacks a programming interface. Anthropic (搜索) will work to safeguard the system from making unforced errors before offering MHS open source. For now, the research preview will be available to a first group of scientific research labs and advanced manufacturers.
Genentech plans to evaluate Claude (搜索) and MHS to orchestrate broader, end-to-end autonomous workflows in its drug discovery labs, aiming to build an autonomous discovery engine where scientists set the high-level biological intent and AI agents help coordinate the physical pipeline. Carnegie Mellon researchers plan to validate the system with real drug candidates and replace the dye's color signal with readouts that capture actual biological effects, as well as expand MHS support to instruments such as qPCR and microscopes.
Last month, Anthropic (搜索) launched Claude (搜索) Science, an AI "workbench" for researchers involved in drug discovery. Claude Science has over 60 functions built in for areas like genomics, single-cell studies, proteomics, structural biology, and cheminformatics, assisting researchers with tasks like 3D protein structure rendering, analysing genome maps, single-cell RNA sequencing analysis, and CRISPR screen design.