Anthropic Tests Standard for AI Agents to Operate Equipment
The Model Hardware Standard gives AI agents a common interface for laboratory and factory devices, but remains an early research project requiring expert oversight.
Topics
News
- Fired OpenAI Researchers Warn Dismissals Could Chill Safety Work
- ARTEX Developer Closes AI Hacking Tool After Bank Attacks
- AI Redraws Global Race for Quant Talent as Coding Tools Lower Entry Barriers
- AI Dispatch | India Draws Billions Into AI
- AI Bots Now Make Up Half of Traffic to US Fed’s FRED Database
- Musk Blames ‘Oligarchs’ For Delay in Starlink’s India launch
Anthropic has opened a research preview of a technical standard designed to let AI agents operate programmable equipment in laboratories, factories and robotics systems.
The Model Hardware Standard, or MHS, allows agents to control devices including microscopes, liquid handlers and robotic arms through a common interface, removing the need for a separate integration for each machine.
Hardware integration that typically takes weeks or months could sometimes be reduced to hours or minutes, Anthropic said on Thursday, 27 August.
MHS began as a collaboration between Anthropic and the HHMI Janelia Research Campus. It uses standardized software drivers that translate basic commands such as reading a temperature or adjusting a setting into instructions that different machines can understand.
The system can also provide an AI agent with information that may not be obvious from software alone, including a machine’s physical characteristics and operating limits. That allows an agent to coordinate several instruments, monitor results and alter parameters while an experiment is running.
During testing, Anthropic said Claude was able to adjust a laser, observe how the beam moved through a camera and repeatedly refine its actions. The model later converted what it had learned into a deterministic script, allowing the task to run without fresh reasoning at every step.
The standard is model-agnostic, meaning it is not restricted to Claude, and can work with agent systems through protocols including Anthropic’s Model Context Protocol.
Early participants and hardware partners include AWS, Automata, Danaher, Doosan Robotics, QIAGEN, Tecan, Universal Robots, Hugging Face and Raspberry Pi.
The move also raises a more difficult safety question: what happens when increasingly autonomous AI agents are given direct control over machines rather than software alone?
Anthropic acknowledged that current models still have limits in physical and spatial reasoning and require expert oversight. In one test involving protein samples, researchers had to help Claude understand that foaming was a physical problem rather than a software failure.
The company said it will use the research preview to develop additional safety tests and a physical-safety roadmap before making MHS open source.
For Anthropic, the project is an early attempt to build a common operating layer between AI agents and physical machinery. If such standards gain adoption, the next phase of agentic AI may be defined not only by what models can do on a computer, but by which real-world equipment they are allowed to control.
