Anthropic Introduces Model Hardware Standard to Help AI Agents Control Physical Machines

AI company Anthropic has announced a research preview of the Model Hardware Standard (MHS), an interface designed to create a common way for AI agents to communicate with and operate programmable physical machines. The Dario Amodei-led company said the standard can work across laboratory and manufacturing equipment, including microscopes, liquid handlers and robotic arms. MHS aims to reduce the need for separate, time-consuming software integrations whenever new equipment is added to a facility.

Under the proposed system, machines can provide AI agents with information about their capabilities, measurements, configurable settings and safety limits. This could allow an AI agent to coordinate several machines simultaneously and carry out complex workflows, from routine drug-discovery experiments to specialised scientific procedures. Anthropic cited experiments involving Genentech, Carnegie Mellon University and HHMI Janelia, where MHS reportedly reduced equipment integration time and improved workflow efficiency. However, the company acknowledged that AI agents still struggle to understand unexpected physical problems. In one experiment, Claude initially responded to bubbles during liquid handling by restarting the process, which made the issue worse until a human explained the underlying cause. Anthropic plans to make MHS open source after the research-preview phase and said the standard is model-agnostic rather than being restricted to Claude.