OpenAI, PerturbAI and Scripps Research ran thousands of AI agents, each powered by GPT-Rosalind, over a whole-brain CRISPR atlas that records how gene disruptions change cell types across the mouse brain. According to OpenAI, each agent took one gene disruption, analysed the cellular response, searched the scientific literature and proposed an explanation; the individual reports were then merged into a reasoning atlas linking genes, cellular effects, publications and candidate mechanisms. The team estimates that the equivalent human analysis, done one perturbation at a time, would take months to a year. OpenAI says the agents reproduced findings from earlier human analyses and surfaced additional leads, and that their novelty and value ratings matched judgements from blinded human experts. Two examples are given: different effects of Tsc1 and Tsc2 disruption in the cerebellum, and a possible link between cholesterol regulation and neurodegeneration. In a comparison that fixed the model and the compute, a single agent cited 11 references at the largest workload against more than 700 for the distributed setup, and blinded experts preferred the distributed reports 75% of the time. A preprint accompanies the release.
Our takeThe controlled comparison is the useful part: with the same model and compute, one agent covered far less literature as the workload grew (11 citations against more than 700), so the gain came from parallel breadth, not a stronger reasoner. Whether the merged reports hold up is a separate question. Agreement with blinded experts is a proxy for plausibility, and the biological leads still need wet-lab confirmation. The released atlas is best read as a prioritisation tool for deciding which experiments to run next, not as evidence of mechanism.