About BroadBox
BroadBox brings scientific agents into controlled environments where they can form hypotheses, choose experiments, learn from real feedback, and revise what they believe.
We came to BroadBox from different parts of the same problem. Arya's work on scientific sandboxes described controlled environments where an AI agent could decide what to test and learn from the result—not simply answer a static question.
At the same time, our work in laboratory automation made a second possibility tangible: those environments could connect directly to experiments. An agent could propose a hypothesis, request a measurement, receive evidence from the lab, and decide what to do next.
We realized that these ideas belonged together. BroadBox defines a new way to benchmark AI scientists for the era of autoscience: by measuring how well they investigate, adapt, and discover when reality is in the loop.
Each sandbox begins with a scientific objective rather than a benchmark answer key.
Agents choose experiments, document their reasoning, and revise their hypotheses.
Experimental results—not fluency—determine how well the scientific process worked.
The paper
The sandbox paper describes environments in which agents can propose experiments, receive feedback from a sealed oracle, and improve their hypotheses over successive rounds. BroadBox turns that idea into a shared platform for real scientific competitions.
Paper link coming soonPeople
Researchers working across scientific agents, experimental design, and laboratory automation.
Portrait placeholder
