Does Turnitin Detect StealthGPT? Vendor Claims vs. What's Actually Verified

In September 2025, Turnitin rolled out a feature it calls AI bypasser detection, built into its existing AI writing detection rather than sold as a separate product. It's designed to flag text showing the fingerprint of having been run through a humanizer or paraphrasing tool specifically to make AI-generated writing read as more human. Turnitin's chief product officer, Annie Chechitelli, framed the update plainly: "we've updated our software to detect leading AI bypasser modifications so educators can uphold academic integrity against this emerging challenge."

Turnitin's own announcement doesn't name any specific humanizer tool — not StealthGPT, not any competitor — and it doesn't publish an accuracy or detection-rate figure for the feature. It also says the feature is currently limited to English-language text. In other words: Turnitin says it built something aimed at tools in this category generally, with no public number attached and no single product singled out.

StealthGPT is an AI content tool that rewrites AI-generated text to try to reduce detection scores, with multiple modes (Standard, and paid Ninja and Ghost modes) offering different levels of rewriting. The only specific bypass number tied to Turnitin that StealthGPT has published comes from a December 2023 press release, where the company claimed its output achieved "a 0% detection rate on Turnitin" based on what it called "a rigorous comparative analysis recently conducted by the StealthGPT team." StealthGPT's CEO, Jozef Gherman, said the software "goes through rigorous testing to ensure we're always ahead of the AI detectors currently on the market."

Two things are worth being precise about here. First, that 0% figure is StealthGPT's own self-run test, not an outside audit — no sample size or methodology was published alongside it. Second, and more importantly, it's from December 2023 — nearly two years before Turnitin's AI bypasser detection feature existed. It cannot be a test of the system Turnitin actually runs today, because that system didn't exist yet.

A 2026 review from Quetext — itself a competing AI detector and humanizer company, so worth reading with that in mind — tested StealthGPT more recently and described its results against Turnitin specifically as "inconsistent," noting that the paid Ghost mode "performs somewhat better on Turnitin" than the free Standard mode. That's a notably more measured claim than StealthGPT's own 0% figure, though it still comes from an interested party rather than a neutral test.

We couldn't find a rigorous, independent test — from a university, a research group, or anyone without a stake in either company's outcome — that ran StealthGPT's current output through Turnitin's current AI bypasser detection feature and published real results. Until one exists, the honest answer is: Turnitin built a feature aimed at this category of tool and hasn't published a number for it; StealthGPT's only published Turnitin-specific figure predates that feature by almost two years; and a more recent review from a competing company calls the real-world results inconsistent rather than the clean 0% the older marketing claim suggests.