How to Detect AI Cheating in an Accountant SQL/Excel Test

By Pinal Dave | Last updated: 2026-08-03

Quick answer

Accountant SQL/Excel tests probe formula logic and data reconciliation — tasks an LLM can solve instantly without the candidate understanding a single function used. In a sql/excel test specifically, the fastest way to catch AI-assisted cheating is to combine an adaptive follow-up question with real-time monitoring of tab focus, clipboard activity, and timing anomalies — a single generic question almost never surfaces it on its own. CodeSignal has observed technical assessment cheating roughly double year-over-year, from about 16% to 35% of sessions. An analysis of 19,368 interviews by Fabric (July 2025-January 2026) found 38.5% of candidates were flagged for AI-cheating behavior — 48% in software engineering roles specifically — and 61% of flagged candidates still scored above the passing threshold.

The threat model: how candidates cheat in an Accountant sql/excel test

SQL/Excel Test is a timed technical skills test administered through a browser-based SQL editor or spreadsheet, usually unproctored by default. For an accountant, that creates specific openings:

Pasting the Excel test's dataset and prompt into an LLM and copying back the formula it generates without understanding it.
Using AI to reconcile a mock ledger and presenting the output methodology as personal technique.
A more skilled colleague completes the timed test remotely while screen-sharing is faked or delayed.

Observable tells

Complex nested formulas (INDEX/MATCH, SUMPRODUCT, pivot logic) appear fully formed with no visible trial-and-error in the formula bar's edit history.
The candidate can't explain what a specific cell reference does when asked to walk through their own formula.
Task completion time is implausibly fast relative to the complexity of the reconciliation involved.

Interviewer script

Use these lines during the sql/excel test itself — they're designed to force live adaptation, which is the one thing a scripted or AI-generated answer can't do convincingly:

"The timer is running and tab-switch activity is logged for this test."
"You'll be asked to explain one of your own formulas or queries live afterward."
"Partial credit is given for visible working, not just a final answer."

What evidence to capture

The full session recording or screen-activity log for the sql/excel test, timestamped against each question asked.
The specific moment you introduced an adaptive follow-up or changed variable, and the candidate's response to it.
Any telemetry available (tab-focus loss, paste events, gaze pattern, second-device detection) rather than relying on interviewer impression alone.
A short written note immediately after the session while the specific inconsistency is fresh — flags made days later are far harder to substantiate.

Which Neuroxa product covers this

Browser Proctoring is the right tool for an accountant sql/excel test. It runs in the candidate's browser during the test or take-home window, flagging tab-focus loss, suspicious paste events, virtual-machine or second-monitor use, and completion-time anomalies — all reviewable afterward as an evidence trail.

Detection signals for Accountant SQL/Excel Test

Detection Signal	Signal Type	Risk Weight
Complex nested formulas (INDEX/MATCH, SUMPRODUCT, pivot logic) appear fully formed with no…	Behavioral / role-specific	Medium
The candidate can't explain what a specific cell reference does when asked to walk through…	Behavioral / role-specific	Medium
Task completion time is implausibly fast relative to the complexity of the reconciliation …	Behavioral / role-specific	Medium
Tab/window focus lost during the test window	Browser telemetry	High
Paste events containing large blocks of pre-formatted text	Clipboard telemetry	High
Answer submitted far faster than the median completion time	Timing anomaly	Medium
Second monitor or virtual machine detected during the session	Environment / device	High

FAQs

Can AI actually cheat effectively in an accountant sql/excel test?

Yes. CodeSignal has observed technical assessment cheating roughly double year-over-year, from about 16% to 35% of sessions. Accountant-specific tasks in a sql/excel test are structured enough that a large language model can produce a fluent, confident-sounding answer in seconds — the risk isn't a lack of AI capability, it's a lack of verification on the hiring side.

What's the single biggest tell for AI use in an accountant sql/excel test?

The most consistent tell across sql/excel test formats is a mismatch between fluency and adaptability: the candidate produces a polished, complete answer instantly, then can't adjust it when you change one variable or ask them to explain their own reasoning in a different way.

Does Browser Proctoring work for sql/excel tests specifically?

Yes — Browser Proctoring is built for browser-based, asynchronous formats like this one, monitoring tab focus, clipboard activity, and session telemetry throughout the test window.

Should we tell accountant candidates the sql/excel test is monitored?

Yes. Disclosed monitoring is both a legal best practice and a deterrent — Karat's data shows that simply moving toward more verified formats (in-person or proctored) has already pushed candidates away from banned-tool use in droves, precisely because the deterrent works before the test starts.

How many accountant candidates are we likely to flag?

Base rates vary by role and format, but Fabric's dataset puts overall AI-cheating flags at 38.5% across interviews, rising to 48% in software engineering specifically — treat any sql/excel test without monitoring as having a meaningful and likely underestimated exposure.

What evidence should we save if we flag an accountant candidate?

Save the session recording or screen-activity log, timestamped notes on the specific question that triggered the follow-up, and the candidate's live response to your adaptive follow-up question — this combination is what holds up if the candidate disputes the flag.

<details> <summary>FAQPage schema (JSON-LD)</summary>

{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
{"@type": "Question", "name": "Can AI actually cheat effectively in an accountant sql/excel test?", "acceptedAnswer": {"@type": "Answer", "text": "Yes. CodeSignal has observed technical assessment cheating roughly double year-over-year, from about 16% to 35% of sessions. Accountant-specific tasks in a sql/excel test are structured enough that a large language model can produce a fluent, confident-sounding answer in seconds — the risk isn't a lack of AI capability, it's a lack of verification on the hiring side."}},
{"@type": "Question", "name": "What's the single biggest tell for AI use in an accountant sql/excel test?", "acceptedAnswer": {"@type": "Answer", "text": "The most consistent tell across sql/excel test formats is a mismatch between fluency and adaptability: the candidate produces a polished, complete answer instantly, then can't adjust it when you change one variable or ask them to explain their own reasoning in a different way."}},
{"@type": "Question", "name": "Does Browser Proctoring work for sql/excel tests specifically?", "acceptedAnswer": {"@type": "Answer", "text": "Yes — Browser Proctoring is built for browser-based, asynchronous formats like this one, monitoring tab focus, clipboard activity, and session telemetry throughout the test window."}},
{"@type": "Question", "name": "Should we tell accountant candidates the sql/excel test is monitored?", "acceptedAnswer": {"@type": "Answer", "text": "Yes. Disclosed monitoring is both a legal best practice and a deterrent — Karat's data shows that simply moving toward more verified formats (in-person or proctored) has already pushed candidates away from banned-tool use in droves, precisely because the deterrent works before the test starts."}},
{"@type": "Question", "name": "How many accountant candidates are we likely to flag?", "acceptedAnswer": {"@type": "Answer", "text": "Base rates vary by role and format, but Fabric's dataset puts overall AI-cheating flags at 38.5% across interviews, rising to 48% in software engineering specifically — treat any sql/excel test without monitoring as having a meaningful and likely underestimated exposure."}},
{"@type": "Question", "name": "What evidence should we save if we flag an accountant candidate?", "acceptedAnswer": {"@type": "Answer", "text": "Save the session recording or screen-activity log, timestamped notes on the specific question that triggered the follow-up, and the candidate's live response to your adaptive follow-up question — this combination is what holds up if the candidate disputes the flag."}}
  ]
}

</details>

Related guides

Neuroxa.ai provides AI proctoring for hiring teams — Browser Proctoring for assessment and take-home formats, and AI Meeting Proctor for live Teams/Zoom interview rounds.