Anthropic is preparing to launch a new research system that enables software to control scientific and engineering equipment for conducting experiments, marking the company's first tool built for physical-world operations. The developer of the Claude chatbot is planning an initial public offering that could value the company at around $2 trillion. Against this backdrop, the technological breakthrough is helping the company expand its enterprise business, driving revenue growth through new applications in manufacturing, robotics, and pharmaceuticals.

The new tool, called the Model Hardware Standard, allows AI agents to autonomously operate a wide range of code-programmable devices, including precision microscopes, liquid handlers, lasers, and robotic arms. Jonah Koor, who leads Anthropic's scientific partnerships and deployment efforts, said: "The conceptual leap here is advancing Claude agents from analyzing and generating hypotheses in the digital world to executing experiments in a real-world laboratory." For instance, high-end microscopes consist of hundreds of mirrors, motors, and lasers that typically require senior engineers to operate. Koor added: "Now Claude can directly interface with and control each individual mirror."

Anthropic has partnered with a group of laboratories and hardware manufacturers in the biotechnology, robotics, and quantum computing sectors to test the technology, including the Howard Hughes Medical Institute, Carnegie Mellon University, Genentech, and quantum computer company QuEra. Researchers have observed that AI agents can track, adjust, and repeat experiments much like human scientists do. Anthropic scientist Alec Kemeny said that during a neuroscience experiment, he witnessed Claude observing live brain tissue for the first time; he asked the AI to locate the alveus—a specific brain structure that even he was previously unfamiliar with. Kemeny noted: "Claude successfully manipulated various mirrors and lasers to achieve focus, locked onto the target region of the brain, and the neuroscientists present confirmed: 'Yes, that's it.'"

Anthropic stated that it plans to open-source the new technology, but only after completing relevant optimizations to guard against various unforced errors in the system. Anthropic CEO Dario Amodei said that society remains less aware of the risks posed by artificial intelligence in the biological domain compared to areas like cybersecurity. Anthropic and other AI developers have issued warnings that malicious users could potentially exploit large models to create novel pathogens or biological weapons.

In June of this year, Anthropic launched its first product for scientists, Claude Science, with applications including rendering three-dimensional protein structures and drug development. Artificial intelligence is becoming increasingly important in drug discovery efforts at major pharmaceutical companies. Eli Lilly, the manufacturer of the blockbuster weight-loss drug, has already partnered with AI chip maker Nvidia and earlier this year invested in Insilico Medicine, a company focused on AI-driven drug discovery. Meanwhile, Google DeepMind founder Demis Hassabis has stepped back from day-to-day management duties to focus on running another venture he started—the AI drug discovery startup Isomorphic Labs.