How to Stop AI Cheating on College Exams: A Cybersecurity Playbook for Higher Education

What One Professor's Exam Scores Reveal About AI Cheating

Back in my day, joining a fraternity afforded you a “test file” or “test bank” — a collection of past exams, quizzes, and homework assignments saved by members of the fraternity. Just an added bonus of forming a brotherhood, and a little assistance with your classes. Is that any different than using AI? That is another conversation. What is different is scale: AI is available to every student, in every class, in real time — and it is beating the exam controls universities rely on.

I recently talked to a Professor about AI and test taking. The professor teaches both online and in-person. This professor mentioned that most of his online students have to report to a test taking center where the exams are proctored and there are some sort of lockdown controls to ensure the test-taking software is the only thing running while they are taking the test.

The professor did mention that there are students that still achieve 100% or on one occasion he let a student take an exam in a library versus in the test taking center. No harm right? The test is still administered and proctored. On this occasion the student achieved 100%. The professor has mentioned that on occasion where he has suspicions, he has the student(s) retake the exams and they achieve a 30% score.

Prior to the advent of AI, the students typically averaged around 50–60% on this particular subject and post AI, the professor is now seeing more 100% results than he ever has. Not that a perfect score isn’t achievable, but the pattern invites suspicion.

What does this tell us? Students are still finding a way to use AI to circumvent controls. But if I were a student, why push for 100% and just fly under the radar at 95 or 98%, it’s still an “A” grade.

Treat Exams Like Sensitive Data

Where do higher education exams and cybersecurity intersect? We must treat exams like sensitive data (PII, ePHI, CUI, etc.), I know it does not result in loss of revenue, but it’s about the integrity of your university. Like a data breach, you do not want to end up on the front page of any news outlet on how students used AI to ace your exams. You also want prospective students and parents to know that you run a university that does not tolerate cheating with AI. When a student finishes at your school, you know they have completed their degree with integrity. Integrity of universities will override prestige in years to come.

How to Combat AI Cheating on Higher Education Exams

Higher education institutions should treat this as a form of AI Governance and shadow IT. A program must be established using a framework like I discussed in my AI Governance & the Shadow IT Problem – Visibility – Policy – Control – Accountability. Here is what higher education teams should be establishing:

1.      Visibility - Know which AI tools and LLMs are actually being used by students.

2.      Policy – Define how AI can be used in students’ day-to-day activities.

3.      Control – Enforce in real time at point of use not after an exam has been completed.

4.      Accountability – Maintain logs and evidence to demonstrate integrity.

Control and Accountability in Practice

Diving deeper into the aspects of Control and Accountability:

1. For writing-based exams and assignments:

   a.      Use incremental drafts, oral defenses, and mixed format questions.

   b.      AI checkers are no longer reliable to detect writing styles.

   c.      Use version tracking platforms to track version control and change tracking. This will prevent the simple copy and paste directly from LLMs onto a document.

2. Add sophistication to camera monitoring: multi-camera proctoring, motion tracking, and keystroke pattern analysis.

3. Lockdown modes and browser isolation – look for processes running in the background, restrict to only a single authorized application.

4. Intercept and prevent block copy and pasting at the endpoint before it reaches the LLMs or AI tool.

5. Screen recording – record the session so it can be reviewed if there are any questions or doubts.

6. Use DNS filtering and network-level filtering to restrict access to known AI domains and API endpoints at the campus or test proctoring facility.

Cost vs. Integrity: Securing Exams on a Tight Budget

All of these controls come at a cost for universities already strapped for cash as enrollment is down. There are ways to reduce these costs like using an Enterprise Browser like Island Browser that combines an isolated browser, data loss prevention (block copy/paste), zero trust network (ZTN), and auditing, in a single platform. But an enterprise browser is only one layer — you still need camera-based proctoring and sound teaching and assessment design.

Before we go back to the 1960s and break out the blue books again, let’s try to make the digital age work.

Get an AI Exam Integrity Assessment for Your Institution

If your institution is seeing exam results that do not match the classroom, the gap is usually visibility and control, not effort. I help colleges and universities assess where AI can slip through exam controls and build a governance program around it: visibility into the AI tools students are actually using, workable policy, enforcement at the point of use, and the logs to prove integrity when someone asks.

Contact me for an AI exam integrity and governance assessment. We will review your current proctoring, lockdown, and network controls, identify the gaps AI is exploiting, and give you a prioritized roadmap that fits your budget.

Reach out at info@nbtsystems.ai or visit nbtsystems.ai to schedule a no-obligation conversation.