Anthropic wants AI agents to be able to drive lab instruments and factory machines, and it has published the safety rules it thinks should govern that work. Its Model Hardware Standard defines how agents may and may not interact with microscopes, liquid-handling systems, quantum computing hardware, manufacturing equipment, and robot arms.
The company sees the physical world as the next frontier for automation, where agents could connect what they read in papers to what happens at the lab bench. It also concedes the dangers: the same interfaces could assist biological weapons work. Anthropic’s answer is guardrails embedded in the models themselves, which it says will block misuse of the standard.
Before general availability, Anthropic plans to refine the safety rules with a small set of trusted partners rather than release the standard broadly on day one.
Alek Kemeny, the quantum physicist who co-led the work, described the goal as closing the loop between AI’s strengths in literature review and data analysis and the hands-on world of experiments.
