The AI Scientist: towards end-to-end automation of AI research

A pipeline that automates the entire scientific life cycle end to end: it creates research ideas, writes code, runs experiments, plots and analyses data, writes the scientific manuscript, and performs its own peer review. A manuscript it generated passed the first round of peer review at a workshop of a top-tier ML conference.

L4Autonomous iterative research Source Peer-reviewed source Outcome Externally validated (partial)

At a glance

Type

Use case

Workflow level

L4 · Autonomous iterative research

Organisation

Sakana AI (with University of British Columbia, University of Oxford, Vector Institute)

Year

2026

Disciplines

Computer scienceMultidisciplinary

Tasks

Automated peer reviewData analysisExperiment designExperiment execution / lab automationHypothesis generationScientific writing

Entry updated

2026-10-15

Evidence & verification

Source level

Peer-reviewed source
Grounded in a peer-reviewed publication. This describes the source; it does not mean the reported results have been independently reproduced.

Outcome verification

Externally validated (partial)
Part of the reported outcome was checked by an external process - for example, collaborating labs running validation experiments, or a generated manuscript passing blind human peer review. Full independent reproduction has not been demonstrated.

Sources checked

2026-10-15 (UTC)

Peer-reviewed in Nature (published online 2026-03-25, DOI 10.1038/s41586-026-10265-5, resolved 2026-10-15). External element: a manuscript generated by the system passed the first round of peer review for a workshop of a top-tier machine-learning conference (workshop acceptance rate 70% per the paper). The end-to-end pipeline claims themselves remain author-reported; full independent reproduction was not identified.

Source-checked does not mean independently reproduced. See the Method page for how these labels are assigned.

Sources (2)

Workflow & human oversight

Iterative idea generation, code execution, experiments, analysis and manuscript evaluation within a research remit. Humans set objectives and compute limits, supervise execution and must review claims before dissemination; automated peer review is not human oversight.

Workflow level is not an evidence score; higher is not better evidence.

Problem

Individual components of the scientific process have been automated before, but a system that autonomously navigates the whole research life cycle - from conception to publication - had remained out of reach. The authors ask how close end-to-end automation now is.

Approach

The AI Scientist is an agentic system built on modern foundation models. It is evaluated in two settings: a focused mode using human-provided code templates as an initial scaffold for a specific topic, and a template-free open-ended mode that uses agentic search for wider scientific exploration.

Outcomes (as reported by the originating team)

The authors report that generated ideas, execution and presentation were of sufficient quality that one manuscript passed the first round of peer review for a workshop whose acceptance rate was 70%, and that both modes produce diverse ideas which are automatically tested, reported on and evaluated (Nature, 2026). The authors flag risks including taxing review systems and adding noise to the literature.

Limitations

Reported results are from ML-research tasks; the peer-review outcome is a single workshop first-round pass, not a full publication. The system's own internal review step is automated and does not substitute for human review. Not independently reproduced.

Reproducibility

The authors publish code at the linked repository (link verified live 2026-10-15). Reproduction or adoption requires budget for running many idea-to-paper cycles, including experiment compute.

Adoption requirements

Frontier model API access with meaningful spend; ML experiment infrastructure; a human to screen outputs before anything leaves the lab, given the authors' own warnings about review load and literature noise.

Comments

No comments yet.

Add a comment

Comments are held for moderation before appearing. Critical or corrective comments are welcome — especially evidence that an entry overstates what has been verified.