We're building a way for AI models to connect to any device and run real experiments.

Summary

Scientists spend a substantial portion of their working hours troubleshooting and connecting laboratory equipment because individual instruments operate on disparate, proprietary communication protocols. To address this friction, Anthropic is developing a standardized way for artificial intelligence models to interface directly with laboratory devices and execute real-world experiments autonomously.

To test this interface standard, researchers connected Claude to a robotic arm that the model had never previously controlled. The experiment involved placing a canned beverage on a lab workbench and prompting Claude from scratch to pick it up. Because the model had not been given specialized pre-programmed routines for the setup, the researchers monitored the live video feed and sensor readouts to see how Claude would coordinate the physical movement.

Operating through the connection standard, Claude generated the necessary commands to position the robotic arm directly above the target object, lower the gripper, and secure the can. The researcher noted that the integration was built from scratch that day and achieved the manipulation task in a matter of minutes.

The demonstration concludes by emphasizing the broader implications of returning time to researchers across complex fields. Direct AI control over laboratory devices is presented as a means to speed up research cycles in domains such as drug discovery, quantum computing, and fusion energy.