Anthropic has introduced a new research system that enables its artificial intelligence software to directly control scientific and engineering equipment, marking the company’s initial foray into AI tools designed for operation in the physical world. The development is part of Anthropic’s broader strategy to expand its enterprise offerings and increase revenue streams ahead of a planned initial public offering, which could value the company at $2 trillion.
The new platform, named Model Hardware Standard, permits AI agents to autonomously manage devices programmable via computer code, including instruments such as advanced microscopes, liquid handlers, lasers, and robotic arms. According to Jonah Cool, Anthropic’s head of partnerships and deployment for science, this represents a significant shift from the company’s existing Claude AI agents, which previously functioned primarily in digital environments focused on data analysis and hypothesis generation.
“Claude is now able to directly talk to and control all of those individual mirrors,” Cool explained in reference to complex microscopes that involve hundreds of components typically operated by specialized engineers. The technology is designed to enable AI to conduct and adjust experiments independently, replicating tasks usually performed by human researchers.
Anthropic has undertaken pilot testing of the Model Hardware Standard with a select group of laboratories and hardware manufacturers spanning biotech, robotics, and quantum computing sectors. Participating institutions include the Howard Hughes Medical Institute, Carnegie Mellon University, Genentech, and QuEra. In these trials, scientists reported that the AI agents were capable of monitoring experimental conditions, making precise adjustments, and repeating procedures in a manner comparable to human operators.
Anthropic scientist Alek Kemeny recounted an experiment involving live brain tissue where Claude was tasked with locating the alveus, a brain structure unfamiliar to Kemeny. The AI successfully manipulated mirrors and lasers to achieve accurate focus, a result confirmed on-site by attending neuroscientists.
While Anthropic intends to eventually release this new technology as open source, the company is taking precautions to mitigate potential risks, including unforced errors that could arise during autonomous operation. CEO Dario Amodei highlighted that public awareness about the dangers of AI in biological contexts lags behind concerns related to cybersecurity. AI experts have warned about the possibility of malicious actors exploiting such technologies to engineer novel diseases or bioweapons.
Earlier this year, Anthropic unveiled Claude Science, its first AI product explicitly developed for scientific applications, including 3D protein structure rendering and drug discovery. The role of AI in pharmaceutical research is rapidly expanding, with major companies like Eli Lilly partnering with AI hardware firms such as Nvidia and investing in AI-driven drug discovery startups like Insilico Medicine.