Anthropic is testing a new standard to let AI agents control robots and lab hardware

Free daily briefing on global business news.

The Model Hardware Standard aims to cut device integration time from weeks to hours, with partners including Genentech, Carnegie Mellon, and Amazon Web Services

NurPhoto / Getty Images

Anthropic announced a new software standard on Thursday designed to let AI agents discover, communicate with, and operate physical hardware such as robotic arms, microscopes, and liquid handlers used in scientific research and manufacturing.

The Model Hardware Standard, or MHS, works by introducing a standardized driver that translates between a computer's operating system and a hardware device using simple commands such as "read" and "write," the company said. The driver also stores information about a device's physical characteristics — weight, safety limits, adjustable parameters — that previously existed only in paper manuals or as tacit knowledge held by specialists.

Anthropic said the standard is not tied to any particular AI model and can interface with any device that exposes a programmable control surface. It works alongside existing protocols, including the Model Context Protocol, or MCP, which Anthropic released as open-source software in 2024 to ease communication between AI agents and external services.

"What MCP did for software, MHS will do for the hardware world," Alek Kemeny, a member of technical staff at Anthropic, told Bloomberg.

The standard is initially available to a select group of organizations through a research preview. Anthropic said it intends to make MHS publicly available once the preview concludes, though the company declined to commit to a specific date. The company said it is using the preview to develop safety evaluations and best practices for AI systems operating physical equipment.

Early partners span several industries. Researchers at Genentech used MHS to automate a protein assay procedure coordinated across a liquid handler, a robotic arm, and a plate reader. Scientists at Carnegie Mellon University ran drug-discovery experiments roughly three times faster than before. QuEra Computing, which builds quantum computers, used MHS to enable an AI agent to recover a laser's precise operating frequency without human intervention 99.3% of the time.

Hardware and software companies are also adding MHS support to their products. Amazon $AMZN Web Services will support MHS through its Strands Robots library. Danaher $DHR and Anthropic said they are exploring how MHS-supported capabilities could advance biomedical research. Doosan Robotics, Tecan, Universal Robots, Hugging Face, and Raspberry Pi are among others building or testing MHS integration.

Elizabeth Kelly, head of beneficial deployments at Anthropic, said the standard has applications beyond science. "We built this for science to sort of show the promise of AI, but there's also huge benefits here for enterprise and for industry," Kelly told CNBC.

Anthropic acknowledged current limitations. Because Claude learns about the physical world through text and images, its spatial and physical reasoning requires expert oversight. During Genentech's testing, researchers had to guide Claude to recognize that errors caused by sample foaming were physical failures rather than software bugs, the company said.

Join 500,000+ readers who start their day with Quartz.

By subscribing, you agree to our Terms of Service and Privacy Policy.

Global business news for a smarter world