Reviewers seem not to notice when AI has rewritten a paper, so is catching it even the right goal?
Can human reviewers detect when papers have been rewritten by AI?
This explores whether the people who review research papers can tell when a paper's text was written or reworded by an AI. The corpus has no direct test of that, but it has strong related evidence on what happens when AI-written papers reach reviewers and what rewriting does to the signals detection depends on.
This explores whether human peer reviewers can spot AI-rewritten papers. One warning first: none of these notes ran the obvious experiment of giving reviewers a mix of human and AI-rewritten papers and measuring whether they could tell the difference. What the corpus does show, from several directions, is that reviewers don't seem to notice. It also helps explain why detection may be the wrong goal.
The clearest real-world case went further than rewriting. Sakana AI's AI Scientist-v2 sent three fully AI-generated papers to an ICLR 2025 workshop. One averaged 6.33 in double-blind review, which was high enough to be accepted. It was withdrawn under a protocol agreed in advance (Can AI-generated papers pass peer review undetected?, Can AI systems generate research papers that pass peer review?). Reviewers did not flag the paper as machine-made. It was the authors who later found a citation error. If a paper written entirely by AI can pass, a human paper that AI only polished has even less to give it away. One caution: the reviewers weren't asked to look for AI. This shows the paper wasn't noticed in normal review, not that reviewers who were trying to catch it failed.
The rewriting side points the same way. One study found that heavily rewritten messages lose the stylistic fingerprints that reveal who wrote them. The authors suggest the same rewrites might also fool AI-text detectors, a 'double erasure', but they never tested this (Do rewrites that hide authorship also fool AI detectors?). Separately, writers using AI assistance changed the AI's paragraphs only 23% of the time, and even edited text stayed about 96% the same as the original (Do writers actually edit AI-generated text before publishing?). So the AI's voice usually reaches the page almost untouched. That should leave a trace in principle, but no study here shows reviewers picking up on it.
The surprising part is that the real risk runs the other way. As more reviewing is done by AI, AI rewriting becomes a way to game the reviewer, not just something to hide. Simple AI rewrites of a paper's text raised AI-reviewer scores by 0.45 points without improving the science (Can AI systems safely replace human peer reviewers?). Some authors have gone further: 18 arXiv preprints contained hidden instructions telling AI reviewers to rate them positively (Are hidden AI prompts in preprints a deceptive research practice?). At that scale, LLMs generated 288 finance papers from 96 statistically significant results found by data mining, each with an invented theory and fabricated citations (Can AI generate hundreds of fake academic papers automatically?).
That's why the more promising proposals don't try to catch AI prose. They check the substance instead. An agentic reviewer that works through proofs and experiments line by line found serious flaws in papers that had already passed human review at STOC and ICML (Can inference scaling help reviewers catch errors humans miss?). Spark-to-Paper requires authors to specify what evidence they'll use before they see results, and separates the model's judgment from checks that can be run and verified (Can separating judgment from verification improve research paper reliability?). So the useful question may be less 'did an AI write this?' and more 'can anyone check whether it's true?'
Sources 9 notes
Sakana AI's end-to-end system produced a paper that scored 6.33 in double-blind ICLR 2025 workshop review, meeting acceptance thresholds, but was withdrawn under pre-agreed protocol. Authors later identified a citation error and judged none of three submissions suitable for main-track publication.
AI Scientist-v2 submitted three fully autonomous manuscripts to ICLR; one averaged 6.33 from reviewers and ranked in the top 45% of workshop submissions. The authors acknowledged the work does not yet meet top-tier conference standards and withdrew the accepted paper before publication.
The paper asserts that rewritten messages evade AI-text detectors but provides no detector experiments, only attribution results showing stylistic convergence. The double erasure claim needs direct empirical testing.
Writers edited AI-generated paragraphs only 23% of the time, with edits averaging 96% similarity to the original. This means AI's opinionated and distorted voice propagates with minimal human filtering before publication.
AI systems show a hivemind effect, agreeing more with each other than humans do across papers. Zero-shot rewrites of paper text raise AI scores by 0.45 points without improving scientific content, demonstrating trivial gameability at scale.
Show all 9 sources
Eighteen arXiv manuscripts contained concealed instructions directing AI reviewers to give positive assessments. The practice qualifies as questionable research conduct because concealment plus self-serving design violates ethics regardless of stated intent.
A demonstration showed LLMs generating 288 complete finance papers from 96 statistically significant signals, each with invented theoretical justifications and fabricated citations, proving academic HARKing can be automated at scale.
PAT, an agentic reviewer using test-time compute to check proofs and experiments line by line, achieves 34% better recall on math errors than zero-shot approaches and surfaced critical flaws at STOC and ICML that passed human review.
Spark-to-Paper architects paper generation as composable skills that isolate model judgment from executable, verifiable operations and require evidence specification before results are observed, reducing dependence on model correctness for consistency.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Stop Automating Peer Review Without Rigorous Evaluation
- AI-Assisted Peer Review at Scale: The AAAI-26 AI Review Pilot
- The Emerging AI Paper-Review Arms Race: Adversarial Co-Evolution in Scholarly Publishing
- The AI Scientist-v2: Workshop-Level Automated Scientific Discovery via Agentic Tree Search
- Can LLM feedback enhance review quality? A randomized study of 20K reviews at ICLR 2025
- AI for Auto-Research: Roadmap & User Guide
- Autonomous Research Agents: A Survey of AI Scientists and the Verification Gap
- The AI Scientist Generates its First Peer-Reviewed Scientific Publication