Defend your research with evidence when review feedback turns personal.

Navigating the peer review process can be challenging, especially when faced with unprofessional behavior from a reviewer.

3 min readMachine Learning

Peer review has a purpose, and it is not this. The ICML 2026 process is currently being held hostage by a single reviewer who lowered a paper to a score of 1, ignored a substantive rebuttal, and replaced technical critique with fabricated references and personal insults. This is not a difference of opinion about methodology; it is a breakdown of the system's basic integrity.

The reviewer's behavior is telling in its specifics. They called the authors "close-minded" and "hostile" while relying on mathematically nonsensical proofs. They made baseless accusations about MIT license and anonymity violations, then used aggressive formatting and syntax errors like bolding ending with periods (**.) to create a visual spectacle. Most tellingly, they kept editing their "PS" section, apparently to bait Program Chair attention and tilt the discussion phase. This is not a scholar engaging in debate. This is someone performing authority while abandoning the actual work of evaluation.

The practical takeaway for anyone reading this is simple: you can do everything right and still face this. The authors responded professionally the first time, addressing each weakness with respect. When they pointed out the reviewer's circular reasoning and lack of relevant citations, the reviewer escalated rather than engaged. That is the clearest signal that the problem was never the paper. The problem is that some reviewers treat the process as a personal stage, not a service to the community.

So what do you do when you encounter this? Document everything. Save the original review, the rebuttal, and every edited "PS" with timestamps. Flag the fake references explicitly in your response to the Program Chair, because fabricated citations are not a matter of interpretation. Ask for the review to be discarded on those grounds, not because the score is low, but because the review contains no valid technical substance. And do not waste your energy trying to win over someone who has already decided you are the enemy. Your other two reviewers gave 5s. One was partially resolved with honest engagement. That is your real signal of quality.

The system only works if we refuse to let a single bad actor poison the well. That means treating a demonstrably fraudulent review as a procedural violation, not a personal setback. It means going to the Program Chair with evidence, not emotion. And it means remembering that your job is to defend your research with evidence, not to appease someone who has abandoned the very standards they were supposed to uphold. If the reviewer wants to live in their own mind, let them. Your job is to make sure the Program Chair knows exactly what is happening on the page.

From Machine Learning

We are currently facing an ICML 2026 reviewer who lowered the score to a 1 (Confidence 5) while ignoring our rebuttal and relying on fake references and personal insults like "close-minded" and "hostile." Despite my other reviewers giving 5s, this individual is using mathematically nonsensical proofs and making baseless accusations about MIT license/anonymity violations, all while using aggressive formatting and strange syntax errors (e.g., bolding ending with periods like **.). The reviewer is also constantly editing their "PS" section to bait Program Chair attention and bias the discussion phase. I’ve never seen such unprofessionalism in peer review; has anyone successfully…

Read the original at Machine Learning