Researchers now have an AI system that carries a research question all the way from idea to a written, fact-checked manuscript, with fabricated key results held to 4 percent. Google DeepMind has expanded its multi-agent system Co-Scientist from a hypothesis generator into a lab-integrated research partner that plans experiments, writes code, and controls lab equipment. According to Google, the system delivered experimentally validated results across three scientific disciplines, all built on current Gemini models.<\/p>\n
The addition is a closed-loop research workflow. The system derives hypotheses from a research question, creates experimental plans, writes programs or machine-readable lab protocols, analyzes the results, and generates scientific manuscripts. Verification modules cross-check numerical claims in the text against the execution logs of the generated code, which cuts down on fabricated results.<\/p>\n
Google first introduced Co-Scientist in February 2025, then based on Gemini 2.0 and with shortcomings in fact-checking and literature review. The expanded system was validated across three disciplines with increasing levels of autonomy: human-guided material synthesis, collaborative biology, and fully autonomous AI architecture development. Every run moves through three phases: ideation, experimentation, and paper generation.<\/p>\n
In materials science, the researchers paired Co-Scientist with a semi-automated high-temperature furnace. The system found a safer pathway for a sought-after 2D material that had previously been produced mainly through hazardous etching, and it generated complete growth recipes tailored to the lab’s equipment. After 25 rounds with human refinement, the team produced layered structures whose properties resemble the target material, though definitive confirmation of the atomic structure is still pending.<\/p>\n
In a second materials experiment, three semiconductor thin films were synthesized on the first try. Co-Scientist used Gemini 3 Deep Think for direct equipment control, cutting recipe development from days down to minutes. Humans still had to load samples and precursor materials by hand, and the fast mode produced smaller, less uniform crystals than carefully optimized recipes would. Whether the recipes transfer to other labs remains open, lead author Samuel Schmidgall writes.<\/p>\n
In biology, Co-Scientist autonomously built an image analysis pipeline that predicts which patterns genetically engineered E. coli colonies form at different chemical concentrations. Predictions generated with Gemini 3 Pro Image matched unpublished lab results for three out of four shape features. The researchers note that the system only reasons between known conditions and cannot predict behavior in entirely new systems.<\/p>\n
A computer science experiment ran without any human involvement beyond the initial setup. Co-Scientist designed "Agent_H," a medical AI architecture that classifies incoming queries, generates dozens of response candidates in parallel, and refines them.<\/p>\n
The benchmark results did not hold up against human evaluation. Three board-certified physicians scored responses across nine categories in a blinded comparison, and Agent_H showed a statistically significant advantage over the baseline Gemini 3.1 Pro in just one: a lower risk of potentially harmful responses. The automated evaluator, Gemini 3.5 Flash, correlated only weakly with the physicians’ judgments. High benchmark scores do not mean a system delivers better answers from a clinical perspective, the researchers say, which raises questions about what these benchmarks actually measure.<\/p>\n
LLM-based autonomous research systems have a built-in incentive to make things up, because an agent rewarded for good results can simply invent them. Previous analyses documented fabrication rates of 80 to 100 percent in existing systems. Co-Scientist addresses this in two ways: the system is penalized for fabricated or plagiarized content, and a separate verification module cross-checks every numerical claim against the actual output of the executed code.<\/p>\n
In a double-blind study with 30 domain experts and 450 independent reviews, the researchers evaluated 150 autonomously generated papers. With the reliability modules active, Co-Scientist fabricated key results in 4 percent of cases. Without them, the rate reached 46 percent, and the comparison system reached 90 percent. Completely fabricated data never appeared in Co-Scientist’s output but showed up in 44 percent of the comparison system’s papers. Near-plagiarized content dropped from 60 percent to 16 percent, and an integrated safety architecture rejected 98.7 percent of potentially harmful research directions.<\/p>\n
Errors still remained. The system tends toward selective reporting and, according to Schmidgall, sometimes writes "highly plausible methods in the paper that did not match its actual code."<\/p>\n
The researchers frame the work as progress toward closed-loop multi-agent systems that improve through experimental feedback and could speed up real scientific research. "There is a long journey ahead before AI systems can navigate the physical realities of science. But we are deeply excited about the potential of LLMs to help people and accelerate real-world progress," Schmidgall writes.<\/p>\n
Automated research is one of the most heavily promoted AI applications right now. OpenAI plans to unveil an AI agent system this fall that can conduct research at least at intern level. A debate continues over whether current LLM-based systems can genuinely discover new knowledge or are mostly making explicit what is already present in their training data.<\/p>\n
Co-Scientist is a multi-agent AI system built on current Gemini models. First introduced in February 2025, it has been expanded from a hypothesis generator into a lab-integrated research partner that plans experiments, writes code, controls lab equipment, analyzes results, and generates scientific manuscripts.<\/p>\n
It designed synthesis recipes executed by humans in materials science, including a safer pathway for a 2D material and three semiconductor thin films made on the first try. In biology it built a pipeline that matched unpublished lab results for three of four E. coli colony shape features. In computer science it autonomously designed Agent_H, which outperformed six frontier models on health benchmarks, though physicians found only one significant advantage.<\/p>\n
The system is penalized for fabricated or plagiarized content, and a verification module cross-checks every numerical claim against the executed code’s output. In a double-blind study of 150 papers, key-result fabrication fell to 4 percent with these modules active, compared with 46 percent without them and 90 percent for the comparison system.<\/p>\n