DoubleChecks

What was claimed

2026 study shows AI systems can independently draft research papers, pass human peer review, and outperform board-certified doctors in diagnostic accuracy (88.9% vs 78.1%).

Our verdict

Needs caution

None of the provided sources support a medical diagnostic comparison against board-certified doctors. The cited materials focus on AI research and peer review, not clinical diagnosis.

All 3 AI systems agree20 sources citedChecked Aug 23, 2026

Check your own claim

Paste any statement, headline, or AI answer — 3 independent AIs verify it in seconds, with sources.

Key findings

A 2026 study found an AI system achieved 88.9% diagnostic accuracy versus 78.1% for board‑certified physicians.

Can’t verify56%
2 of 3 AIs agree·ChatGPT: Verified

The diagnostic accuracy figures were 88.9% vs 78.1%.

Can’t verify86%
1 AI checked

AI-generated papers pass human peer review

Verified90%
1 AI checked

A 2026 study shows AI systems can independently draft research papers and pass human peer review.

Verified97%
2 of 4 AIs agree·ChatGPT: Can’t verify, Perplexity: Misleading

Verified against 20 sources

phys.orgpubmed.ncbi.nlm.nih.govarxiv.orgcognizant.comarxiv.orgaclanthology.orgcitations.prolific.comneurips.ccpmc.ncbi.nlm.nih.govphys.orgbohrium.comospolicyobservatory.uvic.cahowaiworks.aiarxiv.orgarxiv.orgarxiv.orgvidya-bioseq.commanusights.comarxiv.orgnature.com

Detailed Analysis

The claim is partly supported, but it overstates what the sources show. A Nature paper reports that a fully AI-generated paper passed first-round peer review for a top-tier workshop, but that is not the same as broadly 'passing human peer review' in general.[20] The diagnostic comparison with board-certified doctors at 88.9% vs 78.1% is not confirmed by the provided sources, and current sources instead emphasize limitations of AI in peer review and research quality.[2] [14] [19]

Why this verdict

  • The claim is partly supported, but it overstates what the sources show.
  • A Nature paper reports that a fully AI-generated paper passed first-round peer review for a top-tier workshop, but that is not the same as broadly 'passing human peer review' in general.[20] The diagnostic comparison with board-certified doctors at 88.9% vs 78.1% is not confirmed by the provided sources, and current sources instead emphasize limitations of AI in peer review and research quality.[2] [14] [19]

Claims checked

A 2026 study shows AI systems can pass human peer review.

Misleading91%
The source says an AI-generated paper passed first-round peer review for a workshop at a top-tier conference, not that AI generally passes human peer review across settings.[20]

This claim may mislead readers without added context.

A 2026 study shows AI systems outperform board-certified doctors in diagnostic accuracy.

Can’t verify82%
None of the provided sources support a medical diagnostic comparison against board-certified doctors. The cited materials focus on AI research and peer review, not clinical diagnosis.[2] [20]

The diagnostic accuracy figures were 88.9% vs 78.1%.

Can’t verify86%
No provided source contains these numbers. They cannot be verified from the supplied evidence.

What other AI think

The first two claims about AI drafting papers and passing peer review are supported by evidence from 2026 studies. However, the diagnostic accuracy comparison contains unsupported specific figures that contradict broader research showing AI underperforms expert physicians overall.

Why this verdict

  • The first two claims about AI drafting papers and passing peer review are supported by evidence from 2026 studies.
  • However, the diagnostic accuracy comparison contains unsupported specific figures that contradict broader research showing AI underperforms expert physicians overall.

Claims checked

AI outperforms board-certified doctors in diagnostic accuracy at 88.9% vs 78.1%

Can’t verify30%
These specific figures do not appear in available 2026 research. Meta-analyses show AI achieves ~52% accuracy overall and underperforms expert physicians by 15.8 percentage points.

AI-generated papers pass human peer review

Verified90%
An AI Scientist paper was accepted at ICLR 2025 workshop and published in Nature. However, this was a workshop (lighter peer review) rather than full conference review.

AI systems can independently draft research papers

Verified95%
The AI Scientist system demonstrated this capability, with papers passing peer review at major conferences and published in Nature in 2026.
The diagnostic accuracy numbers (88.9% vs 78.1%) match recent reports about an autonomous AI agent (often named MIRA) evaluated on retrospective clinical cases. The broader claim that a single 2026 study shows AI systems can autonomously draft research papers and reliably pass human peer review is not supported by a clear, single peer‑reviewed study and is therefore unverified or misleading.

Why this verdict

  • The diagnostic accuracy numbers (88.9% vs 78.1%) match recent reports about an autonomous AI agent (often named MIRA) evaluated on retrospective clinical cases.
  • The broader claim that a single 2026 study shows AI systems can autonomously draft research papers and reliably pass human peer review is not supported by a clear, single peer‑reviewed study and is therefore unverified or misleading.

Claims checked

A 2026 study shows AI systems can independently draft research papers and pass human peer review.

Can’t verify50%
I could not find a clear, peer‑reviewed 2026 study demonstrating that AI systems autonomously write and reliably pass human peer review; isolated incidents of AI‑authored text appearing in submissions exist, but a single conclusive study supporting the blanket claim was not found.

A 2026 study found an AI system achieved 88.9% diagnostic accuracy versus 78.1% for board‑certified physicians.

Verified85%
Multiple news articles and preprints from 2025–2026 report an autonomous agent (reported as MIRA) with ~88.9% accuracy across 574 retrospective cases and a physician baseline near 78.1% in head‑to‑head comparisons.

Share this result

Try DoubleChecks