AI Detection / Humanizers / Comparison
AI humanizers that beat Turnitin’s new bypasser detection are now scarce: in 2026 testing, StealthGPT, Groby and GPT Human all got flagged, while only lighter tools like Easy Essay slipped through. No humanizer can be called reliably Turnitin-proof.
Key Takeaways
- Turnitin added a dedicated “AI bypasser” detection layer in August 2025, aimed at text rewritten by humanizer tools.
- In a Northumbria University test, StealthGPT jumped from 0% to 72% likely-AI under the new detector, Groby 0% to 67%, and GPT Human hit 31% — ETIH.
- Only Easy Essay stayed at 0%; StealthWriter and Refrazy returned ambiguous 1-19% “unknown” scores.
- No humanizer publishes an independently verified pass rate against the bypasser layer specifically — vendor “99% bypass” claims predate it.
- Undetectable AI still posts the strongest independent general bypass rate (~73.4%) and bundles a built-in detector to self-check — refine.so.
- The only dependable safeguard is a pre-submission AI check plus authorship evidence — drafts and version history, not a humanizer alone.
What is Turnitin’s bypasser detection?
Turnitin’s bypasser detection is a layer added to its AI writing detector that flags text rewritten by “humanizer” or paraphrasing tools, not just raw ChatGPT output. Turnitin announced it on August 27, 2025, describing humanizer services as “a new category of cheating providers.”
Before this, the standard humanizer playbook was simple: generate with AI, run it through a rewriter, submit a clean 0% score. The bypasser layer is Turnitin’s attempt to recognise the rewriting patterns those tools leave behind. We covered the launch in depth in Turnitin now detects AI humanizers.
Which AI humanizers still beat Turnitin’s bypasser detection in 2026?
The honest answer: very few, and none with proof. The most-cited public test comes from Tadhg Blommerde, an Assistant Professor at Northumbria University, who ran six tools through Turnitin’s updated detector. The split was stark.
StealthGPT, one of the most popular tools, saw the biggest swing — from 0% to 72% likely-AI. Easy Essay was the only tool that stayed clean. But this is one academic’s test on a small sample, and Turnitin labels every flag simply “AI generated,” so even the “passes” are noisy. Treat it as a warning shot, not a leaderboard.
How did each humanizer score against the new detector?
| Tool | Turnitin bypasser test result | Independent general bypass | Starting price |
|---|---|---|---|
| Easy Essay | Passed (0%) | Not benchmarked | Free tier |
| StealthWriter | Ambiguous (1-19% “unknown”) | Mid-tier | $12/mo |
| Undetectable AI | Not in this test | ~73.4% (strongest) | $9.99/mo |
| WriteHuman | Not in this test | Strong on voice retention | $12/mo |
| GPT Human | Flagged (31%) | Mid-tier | $12/mo |
| StealthGPT | Flagged (72%) | Fast, short-form | $14.99/mo |
| Groby | Flagged (67%) | Not benchmarked | n/a |
Which humanizers are worth testing against Turnitin?
If you are going to test anything, start with the tools that either have the strongest independent record or a built-in way to check yourself before you submit. Run your own text — not the marketing demo — through each one and verify with a detector.
Undetectable AI
Undetectable AI posts the strongest independent general bypass rate in 2026 testing (~73.4%) and, usefully, bundles its own AI detector so you can self-check a rewrite before it ever reaches Turnitin. With 11M+ users and a University writing mode, it is the most defensible first stop. Read our full Undetectable AI review, or try Undetectable AI.
WriteHuman
WriteHuman is the pick when keeping your own voice matters — it tends to preserve individual phrasing better than tools that flatten everything into the same cadence. It was not in the Northumbria test, so pair it with a pre-check. See the WriteHuman review or try WriteHuman.
StealthGPT
Here is where honesty beats hype: StealthGPT was the worst performer in the test, jumping to 72% likely-AI under the bypasser layer. It is still fast and capable on short-form content, but if your target is Turnitin specifically, do not assume the old 0% scores hold. Details in the StealthGPT review; you can try StealthGPT for non-Turnitin work.
GPT Human
GPT Human landed at 31% likely-AI — below StealthGPT but still flagged. That “31%” is exactly the grey zone that gets a paper pulled for review without proving anything. Our GPTHuman review covers its rewrite quality; try GPTHuman if you want to test it yourself.
How does Undetectable AI compare to StealthGPT and WriteHuman?
On the single axis that matters here — surviving Turnitin’s bypasser layer — Undetectable AI has the best independent record and the self-check advantage, StealthGPT has the worst recent test result, and WriteHuman sits in between with the best voice retention but no bypasser-specific data. For a deeper three-way breakdown see Undetectable AI vs WriteHuman vs StealthGPT.
Why do humanizers get caught by the bypasser layer?
Humanizers rewrite for low “perplexity” — smoother, more predictable phrasing — which is the exact signature the bypasser model is now trained to spot. The rewriting itself leaves a fingerprint. That is why a tool can score 0% on the old AI check and 72% on the new one: it is not detecting the original AI, it is detecting the rewrite. For the underlying numbers, see bypass rates after humanization.
How can you actually stay safe if you use a humanizer?
Stop treating a humanizer as the finish line. The durable safeguards are process, not tools: run a pre-submission Turnitin AI check before you hand anything in, keep your drafts and Google Docs version history as authorship proof, and edit in your own voice rather than trusting a one-click rewrite. Even Turnitin’s own March 2026 guidance says its score “should not be used as the sole basis” for an accusation — which cuts both ways. If you want to lower a score legitimately, start with how to lower a Turnitin AI score without humanizer tricks and the Turnitin score threshold guide.
Frequently asked questions
Does any AI humanizer fully beat Turnitin in 2026?
No tool has independently verified that it fully beats Turnitin’s bypasser detection. In the Northumbria test only Easy Essay stayed at 0%, and that is a single small sample. Assume no humanizer is guaranteed safe.
Did Turnitin really start detecting humanizers?
Yes. Turnitin launched dedicated AI bypasser detection on August 27, 2025, specifically to flag text altered by humanizer and paraphrasing tools.
Is StealthGPT still safe for Turnitin?
Not reliably. In the 2026 test StealthGPT rose from 0% to 72% likely-AI under the new detector. It still works for non-Turnitin short-form content, but do not assume old 0% scores hold.
Can Turnitin detect Undetectable AI?
Undetectable AI was not in the Northumbria test, so there is no public bypasser-layer score for it. It has the strongest independent general bypass rate (~73.4%) and a built-in detector to self-check, but no humanizer can be called Turnitin-proof.
Is Undetectable AI better than WriteHuman for Turnitin?
For raw bypass odds, Undetectable AI has the stronger independent record and lets you self-check. WriteHuman is better at preserving your personal voice. Neither has bypasser-specific test data, so verify with a pre-check either way.
What is the safest way to use a humanizer for a Turnitin assignment?
Run your text through a pre-submission AI check, keep drafts and version history as authorship evidence, and edit in your own voice. The process protects you; the tool alone does not.
Will humanizing get me flagged for cheating?
It can. The bypasser layer is designed to catch rewriting patterns, and a flagged score can trigger an academic-integrity review even when you wrote the underlying ideas. Keep evidence of your own authorship.
Methodology: this comparison synthesises a published Northumbria University bypasser test of six tools (reported by ETIH) and 2026 independent humanizer benchmarks. It is not our own lab test; every figure is linked to its source inline.
