Science One Framework: A verifiable autonomous research framework via Chain-of-Evidence

Google Research · 4 weeks ago · 5

Researchers introduced the Science One Framework, an autonomous research system that uses Chain-of-Evidence (CoE) to ensure AI-generated scientific papers are verifiable by maintaining evidence links from claims back to actual sources and code. The framework eliminated phantom references entirely (versus up to 21% hallucination in baseline systems) while achieving state-of-the-art performance on benchmarks including MLE-Bench and Parameter-Golf. This approach treats verifiability as a core architectural requirement rather than a post-hoc check, enabling AI research agents to produce trustworthy outputs without sacrificing problem-solving capability.