How to Detect AI Cheating in a Cloud Architect SQL/Excel Test

Cloud architect SQL/Excel tests usually cover cost-modeling spreadsheets or capacity-planning queries. AI cheating shows up as instantly-perfect formulas or queries pasted from Copilot/ChatGPT rather than built incrementally, and as browser tab-switches to an AI assistant mid-test. Neuroxa's Browser Proctoring locks down the test tab and logs every window-focus change and paste event.

Threat Model	Observable Tell	Confidence
Query/formula pasted whole from ChatGPT or Copilot	Paste event with no prior keystroke history; formula appears fully formed	High
New browser tab opened to an AI assistant during the test	Tab-focus-loss event logged mid-question with no legitimate reference material configured	High
Excel Copilot auto-completing cost-model formulas	Formula bar shows AI-suggested syntax patterns inconsistent with candidate's earlier manual entries	Medium
Screen-sharing the test to a second person for live help	Webcam shows candidate glancing at a second device/phone repeatedly	Medium-High

Interviewer script (proctor setup note to candidate): "This test is browser-locked — no other tabs, no external AI tools. If you need clarification on a question, use the chat box rather than switching windows." Review the flagged event timeline before scoring the submission.

Evidence to capture:

Full keystroke and paste-event log per answer cell/query box
Tab-focus and window-blur timestamps
Webcam snapshot at each flagged event
Time-per-question compared to cohort baseline
Final answer diff against known AI-generated reference solutions

Neuroxa product: Browser Proctoring — locks the test tab, logs paste events and tab switches, and captures webcam snapshots at flagged moments during the SQL/Excel assessment.

FAQs

Isn't pasting normal if candidates wrote the formula elsewhere first? Neuroxa flags paste-without-keystroke-history as a signal, not an automatic fail — reviewers see the full context before deciding.

Can Excel's own AI features trigger false flags? Yes, which is why we recommend disabling Copilot/Ideas in the test environment ahead of time and disclosing that policy to candidates.

What if the candidate has two monitors for legitimate reasons? Browser Proctoring flags tab/window focus loss regardless of monitor count — a second monitor showing your own reference doc is fine if declared upfront; an AI chat window is not.

How long is the evidence retained? Recordings and event logs are retained per your organization's data retention policy, configurable in the Neuroxa dashboard.

Lock down skills tests with Neuroxa.ai Browser Proctoring.