Pangram and Grammarly's Authorship fuel a campus AI arms race

On US college campuses, an arms race has broken out between AI detectors and the students they scan. Some who never touched a chatbot are running their writing through the same detectors before turning it in, dumbing down sentences and switching on tools that record every keystroke, all to prove a human wrote the paper.

Brittany Carr, a long-distance student in Liberty University's social work program, received failing grades on three assignments an AI detector had flagged; her school required a 'Writing with Integrity' class and a written apology. 'How could AI make any of that up?' Carr told NBC News. 'I spoke about my cancer diagnosis and being depressed.'

On the detection side, Pangram now flags 'humanized' text, meaning chatbot output rewritten by tools like Grammarly, Quillbot or Undetectable AI to read more human, with above 90% accuracy across every humanizer it has tested, and can identify which chatbot family produced the text 93% of the time. Turnitin, whose chief product officer Annie Chechitelli said 'the most important question is not so much about detection, it's really about where's the line,' now tracks around 150 humanizer tools; 43 of those drew 33.9 million website visits combined in October.

Grammarly's Authorship, which Superhuman (the company that makes Grammarly) says has produced 5 million reports in the past year, records which sections a student typed, pasted from another source or generated with AI, so students can hand a professor the replay as evidence they wrote the essay themselves. 'We're going to keep track of when you are going to Wikipedia,' said Jenny Maxwell, Superhuman's head of education, describing what the tool records. 'We're going to keep track of when Grammarly is making suggestions and you're taking them, we're going to keep track of how much time you've spent in this paper or how many sessions.'

Some universities have started backing away from the detectors that fed this loop. Yale, Vanderbilt, Johns Hopkins and Indiana University have adopted integrity policies restricting or discouraging faculty from using AI detectors as sole evidence, and students at Yale, Adelphi and the University of Minnesota have sued their schools over wrongful accusations. Kelsey Auman, a University at Buffalo graduate accused across multiple assignments, gathered more than 1,500 signatures on a petition against detection software and started a group chat she called 'Academic Felons for Life.'

Aldan Creo, a UC San Diego graduate student flagged in November after a teaching assistant said his 'step-by-step explanations' read like ChatGPT, now runs every draft through detectors before submitting and sprinkles in misspellings. 'If we write properly,' he said, 'we get accused of being AI, it's absolutely ridiculous.'