On May 19, 2026, Google DeepMind introduced Co-Scientist, a multi-agent AI system for researchers, and announced that access would begin rolling out in the coming weeks through a new experimental tool called Hypothesis Generation.
Built on Gemini, Co-Scientist operates as a team of specialized agents rather than a conventional chatbot: some propose hypotheses, while others critique, rank and refine them. According to Google DeepMind, the system is designed to help scientists identify promising avenues for experiments more quickly and weed out weak ideas at an early stage.
The company also said that findings from the project had been published in Nature and that the approach had already been tested on problems in biology and medicine, including antibiotic resistance, plant immunity and liver fibrosis. If the tool proves effective in practice, AI will increasingly shift from general-purpose assistants toward specialized collaborators for scientific research.
