AAAI reviewers will be working as free annotators for OpenAI (openreview.net)

AAAI 2026, hosted on OpenReview, received a record 30,948 submissions and is piloting a large-scale, AI-assisted peer-review experiment using a “large frontier reasoning model” from OpenAI. The system generates clearly labeled, non-evaluative AI reviews to supplement (not replace) human reviewers, and it helps post-review workflows by assessing reviewer agreement and drafting meta-reviews for senior program committee consideration. All outputs are intended as informational aids; final decisions and recommendations remain fully human. Authors, reviewers, and committee members will provide feedback on the AI outputs and a technical report of the experiment will be published. Technically, the pilot uses a multi-step workflow with custom tools for technical-accuracy checks, literature search, and results verification, integrated into OpenReview’s modular infrastructure. Fine-grained access controls and anonymity/conflict-of-interest protections were extended to govern AI data access, and OpenReview’s tooling also helped detect dual or overlapping submissions across venues. The experiment tests whether LLMs can scale and speed up review tasks while preserving scientific rigor—potentially reducing reviewer workload and improving discussion efficiency—while explicitly measuring limitations and preserving human oversight and transparency.