Instructors are understandably concerned about academic integrity. As a result, many are trying to find ways to discourage students from using AI to commit academic misconduct in online assignments (e.g., quizzes on Canvas), a significant academic integrity concern. However, as instructors implement measures to guard against cheating, they must ensure that they do not inadvertently implement practices that impede accessibility, which harms disabled students.
This page identifies some common ways instructor attempts to prevent AI misuse interfere with accessibility for students with certain disabilities, including students who are blind, who have low vision, light sensitivity, reading-related disabilities, dyslexia, and other disabilities, as well as many students who are neurodivergent.
General Considerations
Most of the approaches documented on this page are based on the expectation that all students will read and interact with content in a particular way (e.g., reading text visually, with the text looking identical to how it shows up to the instructor), and that only AI systems will use or process the content differently. That assumption doesn't account for the ways that many disabled people read and interact with content, that the result is that many "AI traps" are actually traps for anyone (or anything) that interacts with content in a different way than expected. That means that while the approaches are only intended to catch AI systems, disabled people are also caught in the same net, with no distinction between the two.
Hidden Text
The concerning approach: Embedding or hiding text that is not supposed to be read by students, but will be processed by AI platforms and cause them to answer incorrectly. Common approaches include using text that is the same color as the background (e.g., "white-on-white" text) and/or making text extremely small.
The problems:
- Impact on screen reader users. Screen readers (an assistive technology) turn text into audio, and screen readers do not distinguish between or announce font size or color changes by default. As a result, changing the color and/or size of text will not prevent the screen reader from reading the text to users. This means that students who rely on a screen reader will be unable to distinguish between text that is intended by the instructor to be visible to students and text that is intended to be invisible to students.
- Impact of students with custom text settings. Students with certain disabilities often use custom display settings that change the color, font, and size of text to make it easier for them to read. These settings override the color and size that faculty set for text, making all text look the same, regardless of whether faculty embedded text that was intended to be invisible. As a result, students with disabilities using custom text settings will necessarily view different instructions than students who do not.
Images of Text
The concerning approach: Real text is replaced by images or screenshots of text.
The problems:
- Inability to customize text display. Students with certain disabilities may need to change the color, font, and/or size of text in order to read it. Such customizations only work for real text, not images of text. The use of an image of text, rather than text, prevents disabled students from customizing settings, making it more difficult, or even impossible, for them to read the text.
- Interference with screen readers. Students certain disabilities may use a screen reader that visually highlights text as it is read out. This functionality does not work with images of text.
- Impact on blind students. Students who are blind and using a screen reader need text to be fully navigable and controllable with their screen reader. Images of text can interfere with the use of a screen reader.
Incorrect Alternative Text ("Alt Text Poisoning")
The concerning approach: Embedding or providing false information in alternative text for images as a means of detecting AI, because it processes the alternative text, and not the image itself. For example, inaccurately identifying a person in a portrait, giving false information about the contents of a graph, or adding or removing words or providing entirely different text for an image of text.
The problems:
- Impact on screen reader users. Students who rely on a screen reader will only get the alternative text, and will have no way to know that it is incorrect. They are relying on you to provide accurate alternative text (as required by accessibility standards).
- Improper implementation of the test exception. While there is an exception in alternative text rules if "non-text content is a test or exercise that would be invalid if presented in text," that exception is intended for sensory experiences, like tests of hearing or visual skill development, not simply any content that happens to be on a test. Regardless, that exception still requires that "text alternatives at least provide descriptive identification of the non-text content." False or intentionally misleading alternative text is not a descriptive identification of non-text content.
Sarcastic or Fake Instructions
The concerning approach: Sarcastic, joking, or otherwise fake instructions or details are included with the expectation that all students will immediately recognize they aren't intended to be taken seriously and will therefore ignore them, while AI will follow them exactly as written.
The problem:
- Impact on neurodivergent students. Many neurodivergent students may take instructions literally, and may not recognize sarcasm, jokes, or implications. As a result, such instructions will be followed by some students, regardless of an instructor’s intent.
Relying on Identification of Disabled Students or Accommodations
The problematic approach: Accessibility is interfered with using any of the approaches above, and disabled students are expected or told to identify themself or request accommodations to get a different version of the materials that doesn't have broken accessibility.
The problems:
- Changes in needs. Disabled students may need to change their use of assistive technologies without warning. For example, a student may only need to use a screen reader when they're having a disability flare-up, which can occur suddenly (even in the middle of a short quiz).
- Required outing. Requiring students to out themselves presents a barrier and opens up the potential for additional discrimination and differential treatment, and many disabled and neurodivergent students will choose not to identify themselves for a variety of reasons.
- Conflict with rules. One explicit goal of the 2026 federal web accessibility rules is that students should not need to ask for accommodations for content that is covered by web accessibility standards.
Conclusion
It is critical that instructors design materials that provide equal access for students with disabilities. Disabled students must be able to access web-based content on the same basis as students who do not have disabilities, which requires meeting digital accessibility requirements. This approach removes access barriers and helps ensure a welcoming environment for all students.