A wave of new apps marketed on TikTok and YouTube makes it nearly impossible for teachers to tell whether students are writing their own homework or uploading it to AI. The New York Times reports that tools known as humanizers and autotypers have filled the gap that used to be provided by AI-typed homework, and that the same companies that sell recognition software are sometimes the ones helping students get around.
Tools that work on the checks teachers rely on
Humanizers take AI-generated text and rework it so it doesn’t sound robotic or repetitive enough to get noticed, while autowriters solve the timing problem. Instead of a thousand words appearing in a document at once, which might suggest a teacher checking the history of a version, automatic typists slowly release the text over many hours and include fake writing, deletions, and editing to simulate a real writing session.
Apps like Dripwriter and Duey.ai advertise this directly, telling readers that they can drop down and modify something that appears to have been written by themselves. One app, called Typeflo, promised students they could relax and eat a sandwich while producing their essay. It turned out that it was created and advertised by the son of an Emory University professor, who said he did not know how it got into social media and removed it after being contacted.
Even detectors built to catch AI cannot be trusted
GPTZero’s entire pitch hinges on finding AI writing that other tools miss, but the Times found that a marketer paid by the company created a fake qualified teaching assistant on TikTok to promote it to students. The videos walk students through the GPTZero browser extension, showing them how to filter a paper for AI flags before submitting it and showing that the same tool can generate a full paper with citations from scratch.
In response to the report, the founder and CEO of GPTZero, Edward Tian, said that the company has severed the relationship with the advertiser and is reconsidering whether to retain that paper production capability. Grammarly faces a similar conflict, providing identity testing for teachers while also providing humanizer, text generation, and annotation tools in the same environment. That dishonesty is not limited to these two companies.
A report earlier this year revealed how researchers at the University of Florida tested five popular AI text finders and found false positive rates as high as 99.6 percent, with one word tweak beating most of them completely. The results suggest that schools that rely on these tools for disciplinary decisions are operating with much less certainty than they think.
Skipping AI into classrooms may sound like an obvious fix, but given this lack of trust, schools may have no way to enforce it even if they tried. Some teachers argue that’s beside the point though, since students will need these tools when they enter the workforce.