GPTZero recently rolled out one of its biggest updates yet with the release of the 3.15b detection model. This update is designed to catch:
- AI generated text
- AI paraphrased text
- AI humanized text
- Text that was rewritten to bypass detectors
The update is meant to make GPTZero stricter. But the real question is simple.
Does GPTZero actually detect AI humanizers now?
We tested the update across several popular AI humanizers and the results were surprising.
What Changed in GPTZero’s New 3.15b Update?
On December 18, 2025, GPTZero released its new 2025-12-18-base model (Model 3.15b).
According to GPTZero’s official update notes, this release focuses on one core improvement:
🔧 Main Upgrade
- Improved robustness against AI paraphrasers
This update is designed to improve GPTZero’s ability to detect:
- AI-rewritten text
- AI-paraphrased text
- AI-humanized text
However, based on our testing, Model 3.15b also increased false positives.
Test Results: Does GPTZero Detect AI Humanizers?
We tested multiple humanizers using the same text sample. Here are the results after the update.
AI Humanizer Detection Results
| AI Humanizer | Detected by GPTZero | Result |
|---|---|---|
| TwainGPT | No | Pass |
| Humanize AI | Yes | Fail |
| Undetectable AI | Yes | Fail |
| Grubby AI | Yes | Fail |
| QuillBot | Yes | Fail |
| AI Humanize | Yes | Fail |
| Rewritify | Yes | Fail |
Summary
TwainGPT was the only AI humanizer that bypassed GPTZero.
GPTZero’s Update Increased False Positives
Because the model is now more aggressive, it incorrectly identifies more human written text as AI.
Example
We tested a human written sample from 2013, long before any modern AI writing tools existed.
- Previous GPTZero model result: Human
- New GPTZero 3.15b result: 100% AI
- GPTZero also labeled it as paraphrased AI
This was real human writing with no AI involvement. This shows that the new model is far more sensitive and can easily mislabel human writing.
Why this matters
- Students can be wrongly accused
- Professors may trust incorrect results
- Human writing with clean structure is at risk
- Users lose confidence in the detector
When a model becomes too strict, it sacrifices accuracy.
Impact On Students and Academic Use
GPTZero is one of the most widely used detectors in schools. When an update increases false positives, real students face real consequences.
The new model may:
- Mark human essays as AI
- Create confusion for teachers
- Increase complaints
- Cause disputes over academic integrity
The stricter the detector becomes, the more human writing gets incorrectly flagged.
Does GPTZero Detect AI Humanizers Now?
The GPTZero 3.15b update successfully detected six out of seven humanizer tools we tested.
However, the update also introduced a major issue: an increased false positive rate on genuine human writing.
Humanizers GPTZero Detected (Did Not Bypass)
- Humanize AI
- Undetectable AI
- Grubby AI
- QuillBot
- AI Humanize
- Rewritify
Humanizer That Bypassed GPTZero
- TwainGPT
In short, GPTZero 3.15b is stronger at catching most AI humanizers, but less reliable overall, as it now mislabels authentic human text as AI.
Final Verdict
TwainGPT remains the most reliable AI humanizer and continues to bypass GPTZero, even after the latest update.
That said, users should be aware that AI detectors are not fully dependable. As detection models become stricter, false positives on human writing are becoming more frequent across the industry.