Ryne advertises a 99.9% success rate at bypassing AI detectors, and names ten by brand — Turnitin, GPTZero, Content at Scale, Copyleaks, Originality AI, Winston AI, ZeroGPT, Writer, QuillBot and Crossplag. It describes itself as the #1 AI Humanizer and AI Detector Bypass.
Take the number seriously for a moment, because it is worth understanding rather than dismissing. For 99.9% to mean anything it needs a corpus, a date, a version of each detector, a threshold that counts as passing, and a method someone else can repeat. None of those accompany it.
Without them the figure is unfalsifiable in both directions. We cannot show it is wrong. Ryne cannot show it is right. And a detector vendor can invalidate it overnight with a model update that ships without a changelog, which is the ordinary state of this market rather than an edge case.
There is a specific problem with naming Turnitin in that list. Turnitin is sold to institutions, not students, so a vendor cannot run continuous public tests against the live product a university actually uses. Turnitin also announced detection aimed specifically at humanizer tools in August 2025. A success rate published against it is a claim about a moving system the claimant cannot observe.
HumanFlow does not promise to beat any detector — nobody honestly can — and we publish no bypass rates. That costs us in comparisons like this one, because a page with a number on it reads as more useful than a page explaining why the number would be invented. We would rather lose that comparison than win it dishonestly.