NeurIPS 2026 Rejects 18% of Position Paper Submissions Using AI Detection
Key Takeaways
- ▸NeurIPS desk-rejected 178 submissions (18.4%) and flagged 123 (12.7%) for AI policy violations using Pangram's detection model
- ▸Position Paper Track restricts AI to copy-editing; violations trigger desk rejection or evidence-of-authorship requirements
- ▸Conference emphasizes that slick AI-generated text can mask misalignment with author intent, creating reviewer burden and attribution questions
Summary
The NeurIPS 2026 Position Paper Track has taken enforcement action against AI-generated submissions, desk-rejecting 178 papers (18.4%) and flagging 123 more (12.7%) for verification after partnering with Pangram, an AI detection modeling company. The policy requires submissions to be substantially human-written, with AI use restricted to copy-editing and peripheral changes. This represents one of the first large-scale enforcement actions at a major research conference using AI detection technology.
Conference organizers implemented the strict policy out of concern that AI-generated text—despite often being polished and readable—can depart from authors' original intentions and shift verification burden to peer reviewers. The position paper track chairs emphasized that submitting AI-generated text externalizes the cost of verification to reviewers and raises questions about proper attribution of credit. The action reflects broader academic concerns about maintaining peer review integrity as generative AI becomes more prevalent in research workflows.
- Different NeurIPS tracks have different AI policies; organizers stress authors must comply with track-specific requirements
