Google's Science One Framework Fixes AI Research's Fake Citation Crisis
Google Research's Science One Framework uses Chain-of-Evidence to eliminate hallucinated citations and make AI-generated research fully reproducible and verifiable.
- Google Research introduced the Science One Framework, an experimental autonomous research prototype that eliminates hallucinated citations and unreproducible results.
- The framework is built on Chain-of-Evidence (CoE), which requires every claim in an AI-generated paper to be traceable to a grounded evidence source at the time it is produced.
- Baseline autonomous research systems hallucinate up to 21% of references and pass score verification in as few as 42% of papers; Science One achieves 0 phantom references and perfect score verification.
- A companion tool, the CoE Audit, provides four automated integrity checks (score verification, spec violation, reference verification, method-code alignment) applicable to any AI research system.
- On external benchmarks, Science One earned two Gold Medals and two Silver Medals on MLE-Bench Kaggle tasks, and was the only system to submit a valid entry on Parameter-Golf.
- The framework is an experimental prototype, not a production tool; the paper and project site are publicly available.
Autonomous AI research agents have gotten good at producing polished-looking papers. They review literature, run experiments, and write manuscripts that can pass a casual read. Beneath the surface, many of these papers are broken: citations point to papers that don't exist, reported scores can't be reproduced from the provided code, and the methods described often don't match what the code actually does. Google Research is tackling this with the Science One Framework, an experimental autonomous research prototype built around a new concept called Chain-of-Evidence.
A quiet integrity problem across AI-generated research
Across 75 papers spanning five systems and five frontier research tasks, every baseline system exhibits at least one systematic failure: hallucinated reference rates reach 21%, score verification passes in as few as 42% of papers, and method-code alignment ranges from 20% to 80%. These failures share a common root. No existing evaluation protocol audits whether claims are supported by evidence, and no existing autonomous research system is designed to trace claims back to their sources. The Science One Framework is Google Research's answer to that gap.
Chain-of-Evidence: verifiability built in from the start
The core idea is Chain-of-Evidence (CoE), a conceptual framework that defines what makes a research artifact trustworthy, in the same way that ACID properties define what makes a database transaction reliable. Rather than checking for errors after the fact, CoE requires that every claim in a paper carry a recorded evidence chain at the moment it is produced.