NeurIPS desk-rejected 178 papers for being "AI-generated". The detector flagged the track chairs' own papers at 24-69% [N]

NeurIPS's Position Paper Track desk-rejected 178 submissions, or 18.4% of all papers, based on a proprietary AI detector called Pangram. The detector initially flagged 42.7% of submissions as 90-100% AI-generated, requiring adjustments to reduce the flag rate to 12.7%. There was no human review or appeal process for these rejections. The incident highlights concerns about relying on black-box AI for such critical decisions, especially since the detector also flagged the track chairs' own papers at 24-69%.