Assessment Platforms: Proctoring, Integrity, and Anti-Cheat
Online assessments are a permanent fixture post-pandemic. The cheating problem is real, the solutions are imperfect, and the ethics are nuanced. Here's the technology and operational view.
Key takeaways
- Proctoring modes: live (human watching), AI-assisted (AI flagging, human reviewing), record-and-review.
- Integrity signals: camera, screen, keyboard, browser behavior, IP, identity verification.
- AI proctoring is imperfect; over-trust leads to false accusations.
- Item design (good questions) matters as much as proctoring tech.
Modes of proctoring
Live human proctoring
A human watches the student via webcam during the test. Highest assurance, highest cost.
AI-assisted
AI watches the camera and screen, flags suspicious moments. Human reviews flagged segments.
Record-and-review
Video recorded, reviewed later if score is suspicious.
Lockdown browser
Browser locked to test only; can't open other tabs or apps.
Integrity signals
Identity verification
Photo ID + face match at test start.
Camera
Continuous video of student. AI detects: face leaving frame, second person in frame, looking away from screen, phone visible.
Screen
Continuous screen recording. Detects switching to other apps.
Keyboard and mouse
Behavioral biometrics, typing rhythm, mouse patterns. Detect when style changes mid-test.
Browser
Lockdown browser prevents new tabs, copy/paste, screen sharing.
IP and network
Detect unusual locations, VPNs.
What works for high-stakes
For exam-grade integrity (board exams, certification, hiring tests):
- Identity verification at start
- Continuous camera + screen recording
- AI-flagging with human review
- Lockdown browser
- Question randomization
What's enough for low-stakes
For classroom quizzes:
- Skip proctoring; design questions to discourage cheating
- Open-book / open-internet expectations
- Higher cognitive level questions (apply, analyze) less cheatable than memorization
Ethics and false positives
AI proctoring frequently flags innocent behavior, looking away to think, child interrupting, technical glitches. False accusations are real harm.
Mitigation: human review of every flag before action. Clear appeal process. Transparency to students about what's monitored.
Item design matters
Best anti-cheat is question design:
- Open-ended questions
- Application questions (not recall)
- Question pools with randomization
- Time pressure
- Show-your-work requirements
Common pitfalls
AI proctoring without review. False accusations damage students.
Over-collecting biometric data. Privacy concerns, storage liability.
Lockdown browser as sole defense. Cheaters use second devices.
No accessibility. Proctoring tech disadvantages students with disabilities.
What we recommend
Match proctoring intensity to stakes. High-stakes: live + AI + human review. Low-stakes: design good questions instead.
FAQs
Vendors? Mettl, Talview, ProctorU, custom builds.
Privacy? Disclose what's monitored; minimize collection.
Mobile testing? Possible but harder to monitor; not for high-stakes.