AI Detector Considerations for the Classroom
With the rise of large language models (LLMs) like ChatGPT, Claude, and Gemini, companies have introduced AI detection tools that claim to help educators determine when an LLM has been used to complete an assignment. When thinking about using such tools in the classroom, the Hastings Initiative encourages you to consider the following:
1. Reliability
Performance varies across tools, text length, genre, and hybrid texts (Hadra et al. 2026). Leading tools such as Pangram have improved accuracy (Glickenhaus et al. 2026), but even a very low false positive rate adds up when instructors grade hundreds of assignments per semester. At the scale of the whole institution, this compounds dramatically. Genuine student writing that has been lightly edited using AI may also be flagged as fully AI generated. Further, AI detectors can produce vastly different scores on repeated analysis of the same text, meaning educators cannot make consistent or defensible judgments solely based on their outputs (Malik and Amjad 2025).
2. Bias Against Non-Native Speakers
A Stanford study found that many AI detection tools misclassify texts written by non-native English speakers more frequently than text by native English speakers, unfairly penalizing these students (Liang et al. 2023). Pangram claims its tool does not discriminate against non-native English speakers (Glickenhaus et al. 2026), but this has not been independently verified.
3. Equity
Students with access to sophisticated tools can use them to check and revise their writing until it passes detection or turn to "humanizer" tools built to bypass AI detectors (Masrour et al. 2025), privileging those with the financial resources and AI knowledge to evade detection.
4. Atmosphere of Distrust
AI detection-focused approaches can foster an environment of distrust and anxiety in the classroom, undermining educational relationships (Giray et al. 2025).
The Hastings Initiative does not support the use of AI detectors as the primary source of evidence for student cheating. Even the most accurate detectors are not 100% reliable and relying on a flagged result in an academic integrity charge carries risk of liability. If you choose to use AI detectors in your course, we suggest they be the beginning of a conversation with a student, not the end — and one piece of evidence within a broader case.
Further Reading
- "Managing Artificial Intelligence (AI) in Teaching and Learning," Bowdoin College
- "AI-detection tools have made huge leaps forward — how good are they?" Nature
- "AI Detectors Don’t Work. Here’s What to Do Instead." MIT Sloan Technology Services
- "Pros and Cons of AI Detection," University at Albany