AI writing detectors are fueling a new wave of false accusations
AI-generated text detectors, now widely used by teachers and readers, are increasingly leading to false accusations against people who wrote their own work. Several universities and platforms are already dropping or restricting these tools due to their unreliability.
Since ChatGPT and similar tools surged in popularity, schools and publishers have rapidly adopted so-called AI detectors — tools that use their own AI models to guess whether a piece of writing was created by a human or a machine. A survey by the Center for Democracy and Technology found that around 43 percent of US middle and high school teachers regularly used such detectors between 2024 and 2025.
Unlike plagiarism checkers, which compare text against existing databases, AI detectors such as GPTZero, Pangram, and Turnitin's own tool analyze wording, rhythm, and structure to spot patterns supposedly typical of AI-written text. This approach is far more subjective than matching text online, and it can misfire on writing by non-native English speakers.
Real-world consequences
Several cases have already led to serious fallout. Publisher Minotaur dropped a $2 million book deal last month over suspicions that author Jerry Falade used AI, which he denies. French national Thierry Rignol sued Yale after a professor, relying on GPTZero, accused him of using AI on a final exam, resulting in a failing grade and a one-year suspension. Separately, a student at Adelphi University won a lawsuit this past February after facing a similar accusation.
A 2023 Stanford study found that AI detectors more frequently misidentify writing by non-native English speakers as AI-generated. Neurodivergent writers may also face unfair flagging. Even the tool makers themselves, including Turnitin and GPTZero, acknowledge their results aren't fully accurate and shouldn't be the sole basis for disciplinary action. OpenAI shut down its own detector in 2023 due to poor accuracy.
Universities pull back
Yale, Johns Hopkins, Vanderbilt, and Georgetown have already restricted or dropped AI detection tools. MIT explicitly warns that these tools don't work reliably. Instead, schools are encouraging redesigned assignments — breaking work into smaller stages or letting students disclose AI use without penalty.
Meanwhile, platforms like Substack and LinkedIn are building AI-detection features directly into their products, further deepening mutual distrust between readers and writers.
