Anthropic is launching a research system that allows its software to control scientific and engineering devices to conduct experiments, marking its first tool designed to operate in the physical world. The advance comes as the Claude chatbot maker seeks to expand its enterprise business ahead of its planned $2tn initial public offering.

The new tool, called Model Hardware Standard, enables AI agents to autonomously operate devices ranging from complex microscopes to liquid handlers, lasers and robotic arms. Jonah Cool, who leads partnerships for science at Anthropic, said: "The conceptual leap here is going from Claude agents working in the digital world to do experiments in the physical world."

Anthropic has been testing the technology with labs and hardware manufacturers including the Howard Hughes Medical Institute, Carnegie Mellon University, Genentech and Qu Era. Anthropic scientist Alek Kemeny said Claude identified a brain structure called the alveus during a neuroscience experiment, with a neuroscientist confirming: "Yep, that's right."

Anthropic plans to make the technology open source after safeguarding it from errors. Chief executive Dario Amodei warned society underestimates AI's biology risks. In June, Anthropic launched Claude Science for researchers, while Eli Lilly and Sir Demis Hassabis's Isomorphic Labs pursue AI drug development.