AI Scientist Runs Its Own Experiments, No Human Help Needed

A new artificial intelligence system can plan, carry out, and analyze biomedical experiments entirely on its own. The breakthrough, published in the journal *Science* in August 2026, could speed up medical research by removing the need for constant human oversight. The AI agent, described in the paper "Autonomous biomedical research with an artificial intelligence agent," is designed to work in a laboratory setting. Unlike previous tools that only suggest hypotheses or analyze data, this system physically operates lab equipment, adjusts its methods based on results, and repeats the cycle until it reaches a conclusion. Researchers say this is a major step toward fully automated science. The system can handle routine tasks such as preparing samples, running tests, and recording outcomes. More importantly, it makes decisions mid-experiment—like changing a chemical concentration or extending a reaction time—without waiting for a human to intervene. This capability is significant because biomedical experiments are often time-consuming and require careful, repetitive work. By automating the entire process, the AI could allow scientists to run more tests in less time and reduce errors caused by human fatigue. The paper does not claim the AI is ready to replace human scientists. Instead, it positions the system as a powerful assistant that can work around the clock, freeing researchers to focus on bigger questions like study design and interpreting complex results. Experts note that the technology is still in its early stages. The current system works best with clearly defined protocols and may struggle with unexpected equipment failures or novel problems. However, the authors argue that the ability to autonomously close the loop—from hypothesis to experiment to result—is a critical milestone. If scaled, this approach could accelerate drug discovery, personalized medicine, and basic biology. The next step, the authors write, is to make the AI more adaptable to a wider range of lab environments and more resilient to errors. The study appears in the August 2026 issue of *Science*, Volume 393, Issue 6813.