Shadow 06 / Figure 4

Counting, screening, judging

Scroll the table sideways to see every column.

Counting, screening, judging
Actor What is classified Instrument Use Human step Outcome
Pew Research Center web pages (sample) open-weight detector measure (aggregate) method and threshold published population estimate; individual pages misclassified both ways
arXiv submissions moderation screen staff and moderators limit of two new submissions per month; one-year ban for incontrovertible unchecked AI content
NeurIPS 2026 Position Paper Track submissions commercial AI-text detector screen decision routes varied by score band and stated conditions; additional considerations not listed for every group 178 desk-rejected (no appeal under standard circumstances); 123 offered an evidence route (version history)
Turnitin (vendor guidance) student work own AI-writing detector guidance on use human judgement required result "should not be used as the sole basis for adverse actions"
University of Waterloo student work Turnitin AI detection discontinued — feature switched off, Sep 2025
Curtin University student work Turnitin AI detection discontinued — feature disabled from 1 Jan 2026
University of the Free State student work Turnitin AI detection discontinued — discontinued from 1 Jul 2026
New York Supreme Court, Nassau County one disciplinary determination court review of the evidentiary record and process sanction reviewed court determination annulled on case-specific evidentiary and procedural grounds; no general finding on detectors

The same kind of classifier is used to estimate prevalence, to screen submissions and, in some cases, to inform decisions about individuals.

Note
Aggregate estimation and individual adjudication carry different error requirements (ICML 2024, Liang et al.).

SOURCE · Pew Research Center, 20 Aug 2026; arXiv blog, 1 Oct 2026; NeurIPS blog, 2 Jun 2026; Turnitin guide; University of Waterloo; Curtin University; University of the Free State; Matter of Newby v Adelphi Univ., 2026 NY Slip Op 26021