March 30, 2026
AI writes a research paper that passes peer review. To date, the main role of AI in scientific research has been to assist with narrow tasks such as discovering chemical structures, analyzing data or predicting protein shapes. But now, the technology has broken new ground with a fully AI-generated paper passing peer review at a major machine-learning conference workshop.
A new study published in the journal Nature has introduced the world to The AI Scientist, the first artificial intelligence system designed to automate most stages of the research cycle without human intervention. It was developed by Sakana AI, a Tokyo-based company.
"The AI Scientist...creates research ideas, writes code, runs experiments, plots and analyzes data, writes the entire scientific manuscript, and performs its own peer review," wrote the researchers behind the technology.
Replicating a scientist's workflow
According to the paper, The AI Scientist's workflow mimics that of a real human scientist. First, it generates a list of research directions and checks them against literature databases, discarding those that have already been covered. Then comes the experimentation phase, where it writes the code needed to run experiments, either from a template provided by humans or, increasingly, from scratch.
Once the experiments are concluded, the AI drafts a formal scientific paper in LaTeX (a document preparation system). In addition to writing the methodology, results, discussion and everything else a typical paper contains, it can also search for relevant studies to cite. Finally, the system uses an automated reviewer, another AI trained to grade papers based on accuracy, quality and originality.
To see if the system could compete with human researchers, the team submitted three AI-generated papers to a workshop at the 2025 International Conference on Learning Representations (ICLR). Human reviewers were told that some papers might be AI-generated, but they didn't know which ones.
Passing peer review
Out of the three submissions, one paper received enough scores to be accepted. It earned scores of 6, 7 and 6, with an average of 6.33, which isn't stellar but still a significant milestone for AI.
"This achievement demonstrates the growing capacity of AI for making scientific contributions and signifies a potential paradigm shift in how research is conducted," added the Sakana team.
The paper was later withdrawn as part of the experiment's transparency agreement with conference organizers.
Despite the success, the paper's authors are candid about limitations, such as hallucinations (citing non-existent papers or repeating the same figures in different sections), as well as ethical concerns. These include an influx of AI-generated papers that could overwhelm the peer-review system, making it difficult for the work of real scientists to get noticed.
They also caution that the system could be used for unethical experiments or even to inflate a researcher's credentials. Going forward, Sakana suggests that the technology be developed with clear disclosure standards to maintain the integrity of the scientific process.
Sources: