A collaborative team from Harvard Medical School, the Kempner Institute, and the Broad Institute has open-sourced AutoScientists, an AI system designed for scientific discovery. Built on the ClawInstitute platform, AutoScientists simulates decentralized collaboration, allowing multiple sub-agents to exchange peer reviews before using computational resources. This approach enhances exploration and avoids stagnation seen in previous systems. AutoScientists achieved a 74.4% average Leaderboard percentile on the BioML-Bench benchmark, marking an 8.3% improvement over previous records. The system also improved the Spearman correlation coefficient for ACE2-Spike protein binding prediction by 12.5% and achieved a 6.5% improvement on ProteinGym evaluations, setting new benchmarks in protein engineering and genomics.