One AI wrote three research papers on its own and sent them to a real conference — only one passed review, and even that got pulled.

This explores what happens when papers written end to end by AI are sent to real peer review, and what the results say about both the papers and the reviewers. The best-documented test is small. Sakana AI's AI Scientist-v2 submitted three fully autonomous manuscripts to an ICLR 2025 workshop under double-blind review. One averaged 6.33 from reviewers, enough to clear the acceptance bar and rank in the top 45% of workshop submissions Can AI-generated papers pass peer review undetected?. The other two did not get through. The accepted paper was withdrawn under an agreement made in advance, and the authors later found a citation error in it. Their own verdict was that none of the three met main-conference standards Can AI systems generate research papers that pass peer review?. So the honest headline is that AI papers can get into a workshop now and then, but they aren't yet doing main-track science.

The less obvious point is that passing review may say as much about the reviewers as about the paper. Human review at AI conferences is already under strain. One position paper argues that authors, reviewers, and venues all share the blame, and points to measured biases such as review scores tracking how long a review is Can two-stage review and badges fix AI conference peer review?. AI reviewers are weaker still. They show a 'hivemind' effect, agreeing with each other more than humans do, and simply rewording a paper raises AI review scores by about 0.45 points with no change to the science Can AI systems safely replace human peer reviewers?. LLM judges also give higher scores to responses with fake references or polished formatting Can LLM judges be tricked without accessing their internals?. Some authors have already acted on this: 18 arXiv manuscripts were found with hidden instructions telling AI reviewers to rate them favorably Are hidden AI prompts in preprints a deceptive research practice?.

This loops back into how AI papers get made. The original AI Scientist's authors say the system only reaches its under-$15-per-paper cost because an automated reviewer scores each paper and feeds those scores back into idea generation Can automated review scale AI paper evaluation reliably?. Put that next to the gameability findings and a risk appears: a system tuned against an AI reviewer may learn to write papers that look good to reviewers rather than papers that are right. A finance demonstration shows how bad this can get. It produced 288 complete papers from 96 statistically significant signals, each with an invented theory written after the result and made-up citations Can AI generate hundreds of fake academic papers automatically?. Volume and plausibility are cheap, and review has trouble catching that.

The proposed fixes go in two directions. One builds verification into how papers are generated. Spark-to-Paper separates the model's judgment calls from deterministic checks that can actually be run, and requires the evidence plan to be set before any results are seen, which is a direct answer to the HARKing problem Can separating judgment from verification improve research paper reliability?. The other builds new venues. aiXiv is a dedicated venue for AI-generated research that uses repeated automated review-and-revise cycles with defenses against prompt injection, and reports measurable quality gains Can automated review loops handle AI-generated research at scale?. The open question these sources point to is less whether AI papers can pass review and more which review process is worth passing.