AI & Assessment
The Future of Secure Assessment in Higher Education: AI, Exam Integrity and Independent Capability
Secure assessment should govern the complete environment, support authentic professional tools, and verify what students can do independently.
Secure assessment should verify capability, not merely make cheating difficult. In an AI-enabled environment, that means deciding which tools a student legitimately needs, governing the complete workspace, and preserving enough evidence to trust what the assessment measures.
AI has changed what it means to run a secure exam.
For years, the basic model was relatively straightforward.
Students took a test.
Institutions tried to limit outside resources.
A lockdown browser restricted tabs or applications.
Remote proctoring added cameras, monitoring, or human oversight.
That model worked reasonably well when the main question was whether a student could access unauthorized notes, websites, or another person during an exam.
Generative AI changes the problem.
Students now have access to tools that can write, calculate, code, analyze, summarize, translate, explain, and reason through increasingly complex tasks. Those capabilities are also becoming part of the professional environments students are preparing to enter.
- That creates a difficult question for higher education:
When should students be allowed to use AI, and when does the institution need to verify what they can do independently?
The answer cannot simply be to ban all technology.
It also cannot be to allow every tool during every assessment.
- The future of secure assessment will require something more flexible:
controlled environments that let faculty decide exactly which applications, files, websites, data, and AI tools belong in the exam.
That is the lens we use for this guide.
This article focuses specifically on secure assessment. For the broader institutional picture across AI tutoring, grading, academic integrity, and assessment:
AI Changed the Exam, but Assessment Does Not Have to Move Backward
One response to generative AI has been a return to older assessment methods.
More in-person exams.
More paper.
More blue books.
More traditional proctoring.
There are good reasons institutions are reconsidering how they verify student capability.
But moving backward technologically is not the only option.
- The more useful question is:
What exactly are we trying to measure?
If the learning objective is independent recall or reasoning, then AI may need to be restricted.
If the learning objective is whether a student can analyze a dataset using modern tools, then preventing access to those tools may actually reduce the validity of the assessment.
If the student is being prepared for a profession where AI is part of normal practice, an assessment may reasonably include approved AI.
an environment-level secure assessment model allows faculty to make that decision deliberately.
The environment can be configured around the learning objective rather than forcing every assessment into the same security model.
Independent Capability Still Matters
An AI-first university should not become an AI-dependent university.
Students need to learn how to work effectively with AI.
They also need to demonstrate what they can do without it.
Both matter.
A nursing student may need to understand how AI can support clinical workflows while still demonstrating independent clinical reasoning.
A programmer should know how to use AI coding assistants but still understand the code well enough to debug, explain, and defend it.
An engineering student may use sophisticated software throughout the semester but still need to demonstrate that they understand the underlying principles.
That is why controlled assessment remains important.
The goal is not simply to prevent cheating.
- It is to preserve the institution's ability to say:
This student can actually do this.
That assurance is central to the credibility of a degree.
Practical Assessment Is Different From Traditional Online Testing
One of the strongest use cases for an environment-level secure assessment model is practical assessment.
Traditional online exams often force instructors to simplify tasks because the testing environment cannot support the tools students actually use in the discipline.
an environment-level secure assessment model allows assessments to include real professional applications and resources.
- Current an environment-level secure assessment model examples include environments for:
- Coding and Computer Science
- Students can write, run, test, and debug code using environments such as:
- VS Code
- Eclipse
- PyCharm
- IntelliJ IDEA
- terminals
- supporting files and approved network resources
- Data Science and Analytics
- Assessments can include:
- Jupyter
- RStudio
- MATLAB
- Stata
- datasets
- spreadsheets
- approved documentation
- Engineering and Technical Programs
- Students can work with applications such as:
- SolidWorks
- AutoCAD
- ArcGIS
- Ansys
- Multisim
- Open-Resource and AI-Enabled Assessment
- Faculty can approve specific:
- websites
- reference files
- documentation
- AI assistants
- data sources
- applications
while keeping everything else unavailable.
That allows institutions to assess authentic capability instead of forcing every exam into a multiple-choice format simply because it is easier to secure.
Faculty-Defined Control
Exam security should begin with the faculty member defining the rules.
an environment-level secure assessment model allows institutions to configure the complete workspace around those rules.
Faculty can determine which resources belong in the exam and then apply policies around capabilities such as:
- clipboard use
- file transfer
- printing
- screen sharing
- external storage
- network access
- peripherals
- AI tools
Configurations can also be saved as reusable templates for courses, sections, or future testing periods.
That moves assessment security closer to the academic objective.
- Instead of asking:
How restrictive can we make this exam?
- Faculty can ask:
What does the student legitimately need in order to demonstrate this skill?
Then everything else can remain unavailable.
Browser-Based Does Not Have to Mean Browser-Limited
One important distinction with an environment-level secure assessment model is that students access the environment through a browser, but the exam itself does not have to be limited to web applications.
The computing environment runs remotely.
That means students can launch a standardized workspace from institution-managed or personally owned devices without needing the required professional applications installed locally.
That can also reduce the BYOD problem.
A student's laptop may not be capable of running engineering or data-science software locally.
In a cloud-delivered environment, the application runs remotely rather than depending entirely on the student's hardware.
Every student can therefore receive the same configured environment.
That has implications not only for security, but also for consistency and access.
Secure Assessment Without Making Surveillance the Entire Strategy
Remote proctoring has traditionally approached exam integrity through observation.
Cameras.
Room scans.
Screen monitoring.
Browser restrictions.
Human proctors.
Automated behavioral flags.
Those methods can still be appropriate for certain assessments.
But they are not the only possible security model.
an environment-level secure assessment model focuses primarily on controlling the environment.
In the connected current comparison with ProctorU, ProctorU is characterized around live monitoring, identity verification, screen sharing, and browser controls, while an environment-level secure assessment model emphasizes isolated virtual desktops and restricting the complete workspace to approved resources.
- The distinction is:
- monitor the student
- versus
control the environment.
Those approaches can also coexist depending on institutional requirements.
But for many practical exams, environmental control can reduce the need to treat surveillance as the primary source of security.
Session Activity and Review
Security should also produce evidence institutions can review.
an environment-level secure assessment model can provide authorized faculty or institutional teams with access to relevant session activity and timestamped policy events.
- Examples can include:
- exam session started
- unauthorized site blocked
- clipboard action blocked
- approved AI tool accessed
That allows faculty to understand what occurred during the assessment without relying only on a final pass/fail security indicator.
This also connects directly with the broader process-evidence approach used by a process-evidence model.
What Should Universities Ask Before Choosing a Secure Assessment Platform?
What are we actually trying to measure?
This should be the first question.
Security should support assessment validity, not undermine it.
What tools should students legitimately be allowed to use?
Some assessments should allow no outside resources.
Others may require professional software, documentation, data, or AI.
Does the security model match the assessment?
A multiple-choice quiz and an engineering design exam should not necessarily use the same security model.
Can students use real applications?
If practical capability matters, simulations and browser-only assessments may not be enough.
Does the platform depend on the student's local hardware?
This matters for equity, compatibility, and technical support.
Does the system require invasive monitoring?
Institutions should decide how much surveillance is actually necessary for the assessment risk.
Can faculty review meaningful evidence afterward?
Blocked actions and relevant session events can provide useful context during academic integrity review.
Can the platform support approved AI?
Increasingly, some exams will need to assess whether students can use AI responsibly rather than merely whether they can avoid it.
Secure Assessment Should Verify Capability, Not Just Prevent Cheating
This is the most important distinction.
An exam is not secure merely because cheating is difficult.
A secure assessment should produce trustworthy evidence about what the student can actually do.
Those are related goals, but they are not identical.
If excessive restrictions make the assessment unrealistic, security can actually reduce validity.
If students are expected to use professional applications in their careers, institutions should be able to assess those skills.
If students are expected to work with AI, institutions should eventually assess that capability too.
And when independent reasoning matters, institutions should still be able to remove AI entirely.
The environment should match the objective.
That is the philosophy behind an environment-level secure assessment model.
Open what the student needs.
Close what they do not.
Verify the capability the assessment was designed to measure.
Frequently Asked Questions About Secure Assessment in Higher Education
What is secure assessment?
Secure assessment is the use of policies, technology, environments, or proctoring to ensure that an exam measures the intended student capability under defined conditions.
What is an AI-aware exam?
An AI-aware exam explicitly defines whether AI is permitted and, if so, which tools and types of assistance students can use.
Instead of assuming all AI must be blocked, the assessment rules are aligned with the learning objective.
Can AI be allowed during a secure exam?
Yes.
A controlled environment can permit specific approved AI tools while blocking unauthorized AI or other resources.
an environment-level secure assessment model explicitly supports faculty-approved AI inside configured exam environments.
What is the difference between an environment-level secure assessment model and a lockdown browser?
A traditional lockdown browser primarily restricts the browser session.
an environment-level secure assessment model delivers a complete cloud-hosted workspace that can include applications, datasets, files, websites, network services, and approved AI tools while restricting everything outside the exam policy.
Does an environment-level secure assessment model require software installation?
Students access an environment-level secure assessment model through the browser without installing local exam software, plug-ins, or a VPN.
Can an environment-level secure assessment model run coding or engineering software?
Yes.
the product ecosystem currently lists support for assessment environments using tools such as VS Code, Eclipse, PyCharm, IntelliJ IDEA, Jupyter, RStudio, MATLAB, SolidWorks, AutoCAD, ArcGIS, and other applications.
How does an environment-level secure assessment model connect to academic integrity?
an environment-level secure assessment model controls what students can access during an assessment.
a process-evidence model complements that model by providing process evidence and visibility into how work was created.
How does an environment-level secure assessment model connect to AI tutoring?
a faculty-governed tutoring environment governs AI support while students are learning.
an environment-level secure assessment model allows the institution to independently decide whether that same AI support should be available, restricted, or removed during an assessment.
The Bottom Line
AI has made the old assessment question more important, not less important:
What can this student actually do?
But the answer is not necessarily more paper, more surveillance, or more restrictive technology.
The stronger approach is to define the environment around what the assessment is supposed to measure.
Sometimes that means no AI.
Sometimes it means approved AI.
Sometimes it means Excel, Python, CAD, documentation, or datasets.
Sometimes it means a completely closed environment.
The faculty member should decide.
That is where an environment-level secure assessment model is focused.
Lock the workspace, not just the tab.
Give students the tools the assessment requires.
Remove the tools it does not.
And preserve the institution's ability to verify real student capability.
Verify the capability that matters
Sometimes valid assessment means no AI. Sometimes it means approved AI, professional software, documentation, datasets, or a completely closed environment. Faculty should define that boundary. The environment should match the objective, and the result should preserve the institution's ability to say what a student can actually do.