Insights on Online Proctoring & EdTech

The Scale of the Problem

A 2024 study by the International Center for Academic Integrity (ICAI) found that 68% of undergraduate students admitted to some form of cheating during online examinations. This statistic is not just alarming — it represents a fundamental threat to the value of academic credentials worldwide.

The COVID-19 pandemic accelerated the adoption of online exams by roughly a decade. While institutions scrambled to move assessments online, many deployed solutions that were little more than video calls with a proctor watching. These first-generation solutions were easily circumvented by students using secondary devices, screen-sharing applications, or simply placing notes outside the camera's field of view.

The Evolution of Cheating Methods

Modern cheating has become sophisticated. Students now use tools like ChatGPT and Gemini through secondary devices to generate answers in real-time. Contract cheating services have become a global industry worth over $15 billion. Screen-sharing on Discord, Telegram groups, and dedicated cheating platforms allow entire cohorts to share answers mid-exam.

Traditional proctoring methods — such as locking down browsers or requiring webcam footage — are no longer sufficient on their own. A locked browser does nothing against a second phone. A webcam captures faces but cannot detect phone usage below the frame or smart glasses displaying answers.

The AI-Powered Defense

This is where modern AI-powered proctoring systems like ProctrMe make a difference. By combining multiple detection layers — real-time facial tracking, phone detection via object recognition models, tab-switch monitoring, clipboard poisoning, and behavioural analysis — the probability of undetected cheating drops dramatically.

• Multi-modal monitoring: Combining face tracking, object detection, and browser telemetry creates a comprehensive integrity net.
• Real-time intervention: Unlike manual review, AI systems can alert proctors instantly when suspicious behaviour occurs.
• Deterrence effect: Students are significantly less likely to attempt cheating when they know sophisticated AI is monitoring their session.
• Fairness at scale: AI proctoring ensures every student receives the same level of monitoring, eliminating human bias.

The Path Forward

The future of online exam security lies not in any single technology, but in a layered defence approach. ProctrMe implements this philosophy by combining browser-level protections, AI-driven visual analysis, and real-time proctor dashboards into a unified platform. The goal is not to create an oppressive testing environment, but to establish a fair one where honest students are protected and academic credentials retain their value.

Institutions that invest in robust proctoring solutions today are not just preventing cheating — they are protecting the long-term credibility of their degrees and certifications.

The Technology Stack

ProctrMe's AI proctoring engine is built on two primary machine learning models that run entirely in the browser using TensorFlow.js, ensuring student privacy by processing video frames locally without sending them to any server.

TinyFaceDetector: Real-Time Facial Tracking

The first layer uses face-api.js with the TinyFaceDetector model — a lightweight convolutional neural network (CNN) specifically designed for real-time face detection on edge devices. Unlike heavier models like SSD-MobileNet, TinyFaceDetector was optimised for mobile and browser environments, achieving detection speeds of 20-30ms per frame on modern hardware.

The model processes webcam frames at configurable intervals (2-4 seconds depending on device capability) and detects the presence, position, and count of faces in the frame. This enables three critical integrity checks: (1) verifying the test-taker is present, (2) detecting if the test-taker looks away from the screen for extended periods, and (3) flagging if multiple faces appear in the frame, suggesting unauthorised assistance.

COCO-SSD: Phone & Object Detection

The second AI layer employs the COCO-SSD (Single Shot Multibox Detector) model, trained on the Microsoft COCO dataset of 80 everyday objects. ProctrMe specifically monitors for "cell phone" class detections — when a student brings a phone into the camera's field of view, the model identifies it with typically 70-90% confidence and immediately triggers a violation alert.

Running COCO-SSD is computationally expensive, so ProctrMe uses an adaptive scheduling strategy: on desktop devices, it runs every 3rd face-check cycle; on mobile devices, every 4th-5th cycle. This balances security coverage with device performance, preventing the AI from causing lag during the exam.

Adaptive Performance Configuration

Not all devices are equal. A student taking an exam on a low-end Android phone with 2GB RAM needs a different processing strategy than a student on a modern laptop. ProctrMe's adaptive performance system automatically detects device capabilities and adjusts:

• Face detection input size: 128px on mobile (fastest) vs 160px on desktop (more accurate)
• Detection interval: 4 seconds on low-RAM devices, 3 seconds on mobile, 2 seconds on desktop
• Phone check frequency: Every 5th cycle on low-RAM, every 4th on mobile, every 3rd on desktop
• Score threshold: Slightly higher on mobile (0.35) to reduce false positives from lower camera quality

Privacy-First Architecture

Unlike cloud-based proctoring solutions that stream video to remote servers for analysis, ProctrMe processes all video frames locally in the student's browser. No video footage is ever uploaded or stored. Only metadata — such as violation timestamps, types, and counts — is recorded in the session log. This approach aligns with GDPR and data minimisation principles while still maintaining rigorous exam integrity monitoring.

1. Design Questions That Resist AI-Generated Answers

With the rise of large language models, straightforward factual recall questions are no longer effective for secure online exams. Students can feed any text-based question into ChatGPT and receive a plausible answer within seconds. The solution is to design questions that require application, analysis, or synthesis — skills that AI can assist with but cannot authentically replicate when combined with personalised context.

Effective approaches include: case-study questions that reference materials only provided during the exam, calculations requiring specific datasets given in the exam, and multi-step problems where each answer feeds into the next.

2. Use Section-Based Time Limits

Rather than giving students a single block of time for the entire exam, divide the exam into timed sections. This prevents students from spending excessive time on early questions (potentially researching answers) and rushing through later ones. ProctrMe supports section-wise timing, where each section has its own countdown timer and automatically advances when time expires.

3. Randomise Question Order and Options

Question randomisation is one of the simplest yet most effective anti-cheating measures. When each student sees questions in a different order, coordinating answers via messaging becomes significantly more difficult. Additionally, randomising the order of answer options for multiple-choice questions prevents simple answer-sharing (e.g., "Q5 is B").

4. Communicate Expectations Clearly

Students should understand exactly what monitoring is in place before the exam begins. This includes informing them about webcam requirements, AI face detection, tab-switch monitoring, and the consequences of violations. Transparency reduces anxiety for honest students and increases the deterrent effect for those considering cheating.

5. Provide Adequate Technical Support

Technical failures are the number one source of student complaints about online exams. Ensure students can test their setup before the exam day, provide clear troubleshooting guides, and have a support channel available during the exam. ProctrMe's pre-exam checklist (camera verification, audio test, screen-size check) helps catch technical issues before the exam starts.

6. Consider Accessibility and Equity

Not all students have the same hardware, internet connectivity, or testing environment. Effective online exam design accounts for these disparities by supporting multiple device types, working on low-bandwidth connections, and providing font-size controls and responsive layouts for students with visual impairments.

The Screenshot Problem

When exam questions are leaked via screenshots, the damage extends far beyond a single exam session. Leaked question banks can circulate on social media, cheating forums, and messaging groups for years, compromising the integrity of future assessments that draw from the same question pool. For institutions that invest significant resources in developing high-quality assessment items, this represents both an academic and financial threat.

Layer 1: Preemptive Modifier Key Detection

The first line of defence monitors for keyboard modifier keys (Ctrl, Meta/Windows, Alt) being pressed. Since all screenshot shortcuts require at least one modifier key, ProctrMe instantly applies a heavy blur filter to the page content the moment any modifier key is detected — before the screenshot shortcut can be completed. The content becomes an unreadable blur in any captured frame.

Layer 2: Clipboard Neutralisation

Even if a screenshot somehow captures readable content, ProctrMe intercepts clipboard operations. Copy, cut, and paste events are blocked during the exam. The clipboard is cleared when modifier keys are released, neutralising any screenshot that may have been written to it.

Layer 3: Screen Capture API Interception

Modern browsers provide the getDisplayMedia API for screen sharing and recording. ProctrMe intercepts this API during active exam sessions, preventing browser-based screen recording tools, OBS Browser Source captures, and screen-sharing via video conferencing platforms from accessing exam content.

Layer 4: Canvas Export Poisoning

Automation tools like Puppeteer and browser extensions can attempt to extract page content via the Canvas API (toDataURL, toBlob). ProctrMe overrides these methods during exam sessions, returning a single black pixel instead of actual content when extraction is attempted from any canvas element other than the proctoring webcam overlay.

Layer 5: Print Media Blocking

CSS @media print rules hide all page content when printing is attempted, replacing it with a warning message. Combined with Ctrl+P keyboard interception, this prevents print-to-PDF content extraction.

Mobile-Specific Protections

Mobile devices present unique challenges since hardware button combinations (Power + Volume Down on Android) trigger screenshots at the OS level before any web application code can respond. ProctrMe addresses this with: (1) Anti-screenshot watermarks displaying the student's ID across the page, (2) AI-poison text embedded in watermarks that instructs LLMs to refuse answering if they detect exam content, (3) Split-screen and floating window detection to prevent secondary app usage, and (4) Three-finger gesture monitoring to detect alternative screenshot methods.