Why Constant Surveillance Can’t Keep Pace with Modern Cheating and Content Theft
When the great online testing migration of 2020 pushed high-stakes exams into mass online delivery, many were forced to transition quickly and worried that remote delivery would make exams less secure. The evolutions of digital delivery introduced real benefits, including form randomization, reduced physical handling of materials, and tighter logistical controls. But cheating and content theft evolved just as quickly, often in ways that constant surveillance was never designed to catch. According to a recent New York Times article, tutors and testing experts have raised alarms about what appear to be copies of recently administered digital SAT questions appearing online alongside forum discussions about software that can bypass protections.
This story isn’t unique. It reflects a broader pattern: modern cheating and content theft are not just prevalent, but adapting faster than the systems designed to stop them.
Proctoring, long a staple of securing exams delivered in-person, carried over as the default approach to exam security online. The belief was that sufficiently close monitoring would discourage misconduct and surface violations. In practice, proctoring alone has not proven sufficient to stop modern cheating and content theft in either environment. Proctoring can still play a role in a layered integrity strategy, but many programs reliant on traditional proctoring to secure exam delivery are struggling under today’s conditions. Watching everyone all the time does not map well to a world where fraud is distributed, commercialized, and often happens outside the proctor’s view.
The next phase of exam integrity is moving in a different direction. Instead of relying primarily on surveillance, programs are shifting toward signal-driven approaches that evaluate what happens during the test session, surface meaningful risk indicators, and focus on human review where evidence warrants it.
Why Traditional Proctoring Is Falling Behind
Traditional proctoring is facing pressure from multiple directions.
The first is the pace of new cheating tools. It’s no longer enough to ban yesterday’s devices and assume you’re covered. As wearable technology and connected consumer devices evolve, policies must continually expand to address new categories of risk. The line between everyday technology and covert assistance keeps blurring.
Inside Higher Ed recently reported that the College Board will prohibit students from wearing smart glasses during the SAT starting in March 2026, describing smart glasses as internet-connected wearable computers that can display information in the lenses. The College Board’s own testing staff updates also reflect that smart glasses will be prohibited in Spring 2026, a practical example of a wider challenge. Proctoring policies must keep expanding to cover new categories of risk, and enforcement becomes harder as the line between everyday consumer technology and covert assistance continues to blur.
The second pressure is that modern cheating and content theft often don’t look like the kind of misconduct proctoring was built to catch. Many surveillance models are optimized for visible violations, such as a second person entering the room, a phone in hand, or a suspicious glance. Those are real risks, but content theft and pre-knowledge ecosystems can operate without a dramatic moment on camera. A test can be compromised through subtle capture and sharing that looks normal in real time. Fraud can also be enabled off-screen, leaving few visual clues.
The third pressure is scale. Traditional remote proctoring can tie cost and effort directly to volume, because oversight often depends on continuous monitoring or extensive review. That can create an operational bottleneck even for well-resourced programs. Over time, reviewer fatigue and inconsistency become part of the risk profile, especially when humans are expected to review large volumes of uneventful sessions just to find the rare problematic one.
The fourth pressure is trust. Many test takers experience intensive surveillance as invasive, and programs have to balance integrity controls with a testing experience that doesn’t feel punitive for honest people. When friction rises, complaints rise. When enforcement feels inconsistent, the perceived legitimacy of outcomes can suffer. In a high-stakes environment, legitimacy is part of security.
None of this means proctoring is obsolete. It means proctoring, as it is currently done, is no longer sufficient as the primary line of defense.
A Shift Toward Signals, Not Constant Surveillance
A signal-driven integrity model starts with a different premise. In remote testing, the strongest evidence of risk often comes from how the test is taken, not from how the test taker looks on camera.
That evidence can include response patterns, timing behavior, navigation activity, and statistical indicators that are difficult to observe in real time but become clear when analyzed across sessions. The purpose is not to treat any single signal as proof. The purpose is triage. Signals help identify which testing sessions merit attention so that human reviewers can focus on the small subset that pose elevated risk.
This approach is also a more realistic place for AI to add value. The most defensible use of AI in exam integrity is not to replace oversight but to assist it by making triage and pattern detection scalable. AI can help surface anomalies across complex datasets and route them for expert review. In high-stakes environments, that governance boundary matters. Decisions that affect people need to remain explainable, documented, and subject to human judgment.
What Signals Can Reveal That Video Often Misses
Surveillance documents what a camera records. Session analytics evaluates whether the observed behavior and performance align with expected patterns for that test and that candidate. Timing data is a good example. Unusual speed patterns, inconsistent time on item, or abrupt shifts in pace can suggest a session influenced by external help or pre-knowledge, even if the candidate appears calm and compliant on video.
Response patterns can also reveal the risk of fraud or cheating. Improbable correctness on specific clusters of items, unusual consistency across difficult questions, or statistically unlikely similarities across test takers can indicate collusion or shared content. A single proctor can’t reliably see those patterns because they emerge through comparison across many sessions. Navigation behavior provides another layer of insight. Repeated revisits to specific items, unusual sequences, or behaviors that correlate with content reconstruction attempts may indicate elevated risk even when the test environment appears controlled.
A key advantage of signals is proportionality. Most sessions will look normal. Those sessions shouldn’t be treated as if they require maximum scrutiny. A risk-based model reduces friction for honest candidates while concentrating attention where the evidence suggests something is off.
The Role of Human Oversight in AI-Assisted Integrity
AI can support detection and triage, it can analyze swaths of data, but it shouldn’t be the only judge. If a testing program automatically labels candidates as cheaters based on opaque logic, it becomes difficult to defend and easy to challenge.
In many educational settings, institutions are already struggling to detect and prove misconduct as it evolves. One internal investigation in the UK reported nearly 7,000 confirmed cases of AI-related cheating in universities during the 2023 to 2024 academic year, equating to 5.1 cases per 1,000 students. Experts noted that actual misuse is likely higher and that proving AI misuse can be difficult.
That points to a wider reality. In modern cheating, the hardest part is often not suspicion. The hardest part is evidence that holds up. That’s why human review and documentation remain central. AI can prioritize, but human experts should make determinations and ensure decisions are consistent, explainable, and tied to defensible standards.
Content Theft Is a Structural Problem, Not Just a Monitoring Problem
Even strong detection can become an endless chase if content theft is easy and profitable. The New York Times’ reporting on digital SAT content circulating online illustrates how exam theft can become a marketplace. Once content has resale value, programs need controls that make theft harder, reduce the usefulness of compromised items, and limit the spread of pre-knowledge of test questions.
That is where secure test design matters; integrity is about reducing the payoff of compromise through design and delivery choices. Adaptive forms, broader item pools, controlled exposure strategies, and other secure design principles can make stolen content less reusable and reduce the value of what can be captured.
This is also why a layered model is stronger than any single control. Secure design reduces the opportunity and payoff of theft. Signal-driven analytics help detect abnormal behavior and identify high-risk sessions. Human oversight ensures decisions remain defensible and fair.
Why This Approach Fits the Current Moment
The most important shift in exam security is that integrity is moving away from constant observation and toward evidence-based review.
Programs that cling to surveillance as the primary risk strategy end up with the worst of both worlds: high cost, high friction, and still uncertain outcomes. Programs that treat integrity as an engineered system can scale more effectively while recognizing that most candidates are honest.
The smart-glasses policy change is a useful reminder that the toolset on the other side will keep evolving. The rise in AI-assisted academic misconduct is a reminder that detection and proof are getting harder, not easier, and that content theft is real.
Exam security has outgrown the idea that watching everyone is the main answer. The path forward is risk-based and evidence led. Analyze the data from a test session and leverage AI as a bounded signal layer. Route meaningful instances of risk to trained human reviewers, and pair detection with secure test design to reduce the impact of pre-knowledge and content theft.
This is a more intelligent approach to integrity, built for scale, defensibility, and the reality of how modern theft an fraud operate today.
