NeurIPS used uncalibrated AI detector for desk rejections [D]

I recently had a submission desk-rejected from the NeurIPS 2026 Position Paper Track for an alleged AI-policy violation. After corresponding with the track leadership and reading their public blog post, I think the broader methodological issue is worth discussing here. The track used Pangram, a proprietary AI-text detector, as part of the desk-rejection process. I was told that the materials considered for desk rejection were: the detector output the authors’ AI-use attestation This creates a potential circ