Are the fairness risks of AI-assisted homework being underestimated?

How big is the fairness problem—and who is most affected?

The scale of the issue is substantial and growing. A 2026 survey of 1,214 U.S. middle school through college students found that over the course of 2025, increasingly more students used AI for homework, and 60% of those users expressed concern about using AI for school-related purposes [3]. This widespread adoption means that any fairness risk—such as unequal access to AI tools or guidance—affects millions of students. The same survey highlights that students themselves recognize the potential harm to critical thinking, which is a core skill that homework is meant to develop.

The fairness risk is not just about access; it's about outcomes. A 2025 study in an environmental data science course found that while AI tools improved student performance on complex tasks, the benefit was significantly larger when students also received AI guidance—not just the tool itself [2]. This suggests that students who have better instruction or more support at home may gain more from AI, widening the gap between those with and without such resources. The study's authors explicitly note that some students expressed concerns about overreliance, which could disproportionately affect less confident or less supported learners.

How does AI-assisted homework undermine fair assessment?

AI tools can make it nearly impossible to tell whether a student has done their own work, which directly threatens the fairness of grades. A 2023 study tested ChatGPT on university-level courses and found it received near-perfect grades on most questions in introductory courses, though it struggled with advanced material [4]. Critically, the study also found that adding a few typos or full stops could fool machine-learning algorithms designed to detect AI-generated text, meaning even detection systems are unreliable. This creates a situation where some students can effectively cheat without being caught, while others who do their own work may be unfairly graded lower.

This problem is not new, but AI makes it far more scalable. A 2021 study on homework assistance websites found that students performed significantly better on quiz questions with easily searchable answers compared to altered questions that returned no usable results online [5]. When access to those websites was blocked, performance on the searchable questions dropped sharply, while performance on the altered questions stayed the same. This shows that students will use available shortcuts, and AI tools are now the most powerful shortcut yet—available 24/7, free or cheap, and able to generate original-sounding text that evades plagiarism detectors.

What can educators do to reduce the fairness risks?

The evidence points to a balanced approach that integrates AI use into assessment design rather than trying to ban it. The 2025 FACT assessment model—Fundamental, Applied, Conceptual, critical Thinking—offers a concrete example: it combines assignments done without AI (to build foundational skills), AI-assisted projects (to engage with real-world tasks), traditional paper exams (to test independent understanding), and complex AI-assisted case studies (to assess critical thinking) [2]. In this course, students performed better when they had both AI tools and AI guidance, compared to AI tools alone. This suggests that structured, guided use of AI can actually improve learning outcomes while maintaining fairness.

However, the same study also found that some students worried about overreliance, and the 2026 survey confirms that many students share this concern [3]. A 2024 interview study with 20 Chinese university students using a ChatGPT-embedded writing system found that students saw AI as a helpful assistant but also identified challenges related to AI itself, their own skills, and the tasks they were given [1]. These findings together suggest that fairness risks can be reduced—but only if educators actively design assessments that require independent thinking, provide clear guidance on appropriate AI use, and monitor for overreliance. Ignoring the problem or simply banning AI is likely to backfire, as students will use it anyway, and those with better access or guidance will gain an unfair advantage.

About These Sources

This answer is built on 5 peer-reviewed studies — published from 2021 to 2026, 3 from 2024 or later, 2 in Q1 journals, collectively cited 295 times — selected as the most relevant from 5 studies that passed quality screening, drawn from 57 papers retrieved from a database of over 500 million.

Sources used in this answer

Related research questions