Anthropic unveils safety framework for AI agents to operate lab and factory hardware

Anthropic announced a set of guidelines called the Model Hardware Standard, which specifies permissible and prohibited actions for AI agents when using physical hardware. The rules cover a range of equipment—from laboratory microscopes and liquid handlers to manufacturing robots and quantum processors—intended to enable safe automation of experiments and production. Co-lead developer Alek Kemeny, a quantum physicist, said the goal is to close the loop between data analysis and hands-on experimentation, while Jonah Cool highlighted the potential to replace complex engineering tasks with AI-driven configuration.

The company will collaborate with selected partners to validate safety measures before a wider rollout, noting that built-in model guardrails should deter malicious exploitation such as biological weapon creation. Anthropic joins other well-funded ventures like Periodic Labs, LILA Sciences, Edison Scientific and Discovery Loop, all exploring AI-powered scientific discovery. The initiative builds on Anthropic’s earlier Model Context Protocol, which governed software interactions, extending it to the physical realm.