Verifiable Agentic Science on Public Infrastructure
- 1Penn State
Abstract
Autonomous artificial intelligence agents are transforming the practice of science. Across every discipline, researchers are handing data analysis to coding agents that can design and execute complex studies from a single prompt. Yet this revolution carries a threat. When agents improvise in local environments, they leave no verifiable record, ultimately risking a wave of irreproducible science. Solving this crisis requires moving agentic science onto public infrastructure where every action leaves an examinable trail. Here we connect autonomous coding agents to shared computing resources, turning ephemeral prompts into auditable scientific records. After showing that 30 unguided local agents splinter into 18 incompatible pipelines on the same dataset, we conduct an agentic reproduction and discovery study on a large scale---reproducing a large, diverse array of studies with every workflow, history, and intermediate dataset publicly accessible. This complete provenance trail exposes unstated modeling choices, reference mismatches, and subsampled tutorial inputs in published studies, and establishes a new foundation where human scientists can understand, steer, and trust AI-driven discovery.
Related
- MetagenomicsBench: A Verifiable Benchmark for Agentic Metagenomics Analysis
- From FAIR to AI-FAIR Metadata-First Infrastructure for AI-Enabled Science
- Beyond deficit and coexistence: Modeling the knowledge–conspiracy–mistrust configuration in public understanding of science
- Sustainable nutrition: integrating food systems, environmental science, and diet to advance public and planetary health
- S2F-Agent: Harnessing sequence-to-function models for verifiable genome interpretation
BibTeX
@misc{baker2026verifiable,
title = {{Verifiable Agentic Science on Public Infrastructure}},
author = {D. Baker and S. Cain and M. van den Beek and J. Chilton and D. Callan and S. Weaver and N. Coraor and K. Beavers and J. Goecks and M. C. Schatz and B. Gruening and S. L. Kosakovsky Pond and A. Nekrutenko},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.26.761864},
url = {https://www.biorxiv.org/content/10.64898/2026.10.26.761864v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Verifiable Agentic Science on Public Infrastructure
AU - D. Baker
AU - S. Cain
AU - M. van den Beek
AU - J. Chilton
AU - D. Callan
AU - S. Weaver
AU - N. Coraor
AU - K. Beavers
AU - J. Goecks
AU - M. C. Schatz
AU - B. Gruening
AU - S. L. Kosakovsky Pond
AU - A. Nekrutenko
PY - 2026
DA - 2026/10/26
PB - bioRxiv
DO - 10.64898/2026.10.26.761864
UR - https://www.biorxiv.org/content/10.64898/2026.10.26.761864v1
AB - Autonomous artificial intelligence agents are transforming the practice of science. Across every discipline, researchers are handing data analysis to coding agents that can design and execute complex studies from a single prompt. Yet this revolution carries a threat. When agents improvise in local environments, they leave no verifiable record, ultimately risking a wave of irreproducible science. Solving this crisis requires moving agentic science onto public infrastructure where every action leaves an examinable trail. Here we connect autonomous coding agents to shared computing resources, turning ephemeral prompts into auditable scientific records. After showing that 30 unguided local agents splinter into 18 incompatible pipelines on the same dataset, we conduct an agentic reproduction and discovery study on a large scale---reproducing a large, diverse array of studies with every workflow, history, and intermediate dataset publicly accessible. This complete provenance trail exposes unstated modeling choices, reference mismatches, and subsampled tutorial inputs in published studies, and establishes a new foundation where human scientists can understand, steer, and trust AI-driven discovery.
ER -