Google Deepmind's AI Co-Scientist now plans experiments, runs lab equipment, and writes scientific papers

Google Deepmind says its multi-agent system Co-Scientist has delivered experimentally validated results across three disciplines. Expanded from a hypothesis generator first introduced in February 2025 on Gemini 2.0, the system now plans experiments, controls lab equipment, and writes manuscripts. Verification modules cut fabricated key results to 4 percent versus 46 percent without them. Human physician evaluation, however, showed only one significant advantage over the baseline, and atomic-structure confirmation remains pending.