Anthropic launches tool that can manipulate laboratory tools

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 and boost revenues with new applications in manufacturing, robotics and pharmaceuticals, ahead of its planned initial public offering, which could value the company at $2tn.

The new tool, called Model Hardware Standard, enables AI agents to autonomously operate a wide range of devices that can be programmed using computer code, ranging from complex microscopes to liquid handlers, lasers and robotic arms.

“The conceptual leap here is going from Claude agents working in the digital world [of] analysis, hypothesis generation . . . to do experiments in the physical world, in the lab,” said Jonah Cool, who leads partnerships and deployment for science at Anthropic.

For instance, advanced microscopes can be composed of hundreds of mirrors, motors and lasers that are normally operated by expert engineers. “Claude is now able to directly talk to and control all of those individual mirrors,” Cool added.

Anthropic has been testing the technology with a small group of labs and hardware manufacturers in biotech, robotics and quantum computing, including the Howard Hughes Medical Institute, Carnegie Mellon University, Genentech and QuEra, a company that is building quantum computers.

The scientists observed the AI agents tracking, adjusting and repeating experiments in a similar way to human scientists.

Alek Kemeny, an Anthropic scientist, said he observed Claude looking at live brain tissue for the first time during a neuroscience experiment, and asked it to find the alveus, a specific part of the brain that he had not heard of before.

“[Claude] was able to successfully manipulate all the different mirrors and lasers to find the focus on the correct part of the brain, and the neuroscientist sitting there said: ‘Yep, that’s right’,” said Kemeny.

Anthropic says it plans to make this new technology open source, but not until it has worked to safeguard the system from making unforced errors.

Dario Amodei, chief executive of Anthropic, said society lacked awareness of the risks AI poses around biology compared with areas such as cyber security. Anthropic and other AI developers have cautioned that bad actors could use models to create novel diseases or bioweapons.

In June, Anthropic launched Claude Science, its first product dedicated to scientists, with use cases including rendering 3D protein structures and drug discovery.

AI has become increasingly important to drug research at big pharmaceutical companies. Eli Lilly, maker of popular weight-loss drugs, has partnered with AI chipmaker Nvidia chips and earlier this year invested in Insilico Medicine, a company focused on AI for drug discovery.

Meanwhile, Sir Demis Hassabis, founder of Google DeepMind, has left his day-to-day role to focus on leading his other venture, Isomorphic Labs, an AI drug development start-up.

Additional reporting from Cristina Criddle