Anthropic's MHS Lets AI Control Physical Devices
By @sharedot · · 8 pages
- AI
- AI Hardware
- Robotics
- Anthropic
Anthropic previewed a hardware standard letting AI agents operate lab equipment and robots.
Anthropic unveils Model Hardware Standard
On August 27, 2026, Anthropic 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.
Anthropic announced that MHS is being shared with a first group of scientific research labs and advanced manufacturers, and the company plans to eventually open-source the standard after the preview period.
The initiative began as a collaboration between Anthropic and the HHMI Janelia Research Campus.
Why this is a surprising leap
Until now, agentic AI systems have been primarily limited to text, images, code, and other actions inside a computer, as Ars Technica reports.
MHS extends AI control into the physical world, letting models like Claude reason through experimental steps, adjust parameters in real time, and in some cases recover from hardware errors without human intervention.
According to The Cryptonomist, quantum computing firm QuEra used MHS-integrated AI to achieve a 99.3% success rate on laser relock tasks, up from a prior benchmark of 58% with custom scripts.
How MHS works under the hood
MHS introduces a standardized driver that uses simple primitives such as "read" and "write" commands, allowing any compatible device to be discoverable and controllable across a network without bespoke translator software, Anthropic explains.
The driver includes natural-language tags encoding a device's physical characteristics, adjustable parameters, and enforced safety limits, producing a reference file that tells an AI model how to operate equipment it has never seen before.
Agents can control devices through three mechanisms: the Model Context Protocol, a command-line interface, and API code files.
Early partners report dramatic gains
Anthropic says early testing with partners over the past year showed MHS reduced device integration time and enabled faster iteration across experimental settings.
ETV Bharat reports that companies including Amazon Web Services, Automata, Danaher, Doosan Robotics, MBF Bioscience, Tecan, Universal Robots, and QIAGEN are building MHS support into their products.
According to BigGo Finance, Elizabeth Kelly, Anthropic's head of beneficial deployments, said the company built MHS for science to show the promise of AI, but noted huge benefits for enterprise and industry as well.
Safety constraints and current limitations
Safety is embedded directly into the standard, with operational constraints such as speed limits and angle restrictions for robotic systems so an AI agent cannot instruct a device to operate outside safe parameters, The Cryptonomist reports.
Anthropic acknowledges that as a large language model, Claude learns about the physical world through text and images, meaning its spatial and physical reasoning still has limitations requiring expert oversight.
For example, Genentech researchers had to guide Claude to recognize that foaming in protein samples was a physical failure, not a software bug.
What comes next for AI-driven hardware
Anthropic says it will use the research preview period to build additional safety evaluations with launch partners and strengthen protections against misuse before open-sourcing MHS.
The company is developing a physical safety roadmap to bolster safeguards policy and enforcement coverage for AI use in the physical world.
BigGo Finance notes that Anthropic is separately building a silicon team and hiring hardware talent, positioning MHS to compete for the connective layer between AI reasoning and physical execution in a market Barclays projects could reach a trillion-dollar scale by 2035.
Sources
- anthropic.com › Previewing the Model Hardware Standard
- arstechnica.com › Anthropic's new hardware standard lets AI agents control the physical world
- en.cryptonomist.ch › Anthropic AI Hardware Control Advances with Model Hardware Standard
- etvbharat.com › Anthropic Opens Preview Of New Standard Allowing AI Agents To Control Lab Equipment
- finance.biggo.com › Anthropic Debuts Open Standard to Connect AI Assistants With Robots and Lab Equipment