Solo developer's classifier flags over half of ICLR 2027 submissions as AI-assisted

Alice Rigg (@woog09), a solo developer, says she built a classifier to detect LLM-aided writing and ran it on all 42,000+ ICLR 2027 submissions.

Her post says over half the papers appear to be roughly fully AI-assisted. Sentence-level results are hosted at opentextdetector.com in a PDF viewer.

The claim comes from a single X post. No accuracy figures, validation method, or ICLR response appear in the reports, so the classifier's reliability and the finding's accuracy are not established.