If a cheating candidate still passes the interview, why does it matter?
By Pinal Dave | Last updated: 2026-08-01
TL;DR: It matters because a passing score built on AI assistance does not measure the skill the job actually requires. Fabric data shows 61% of software engineering candidates flagged for AI-cheating behavior still scored above the passing threshold, meaning the score itself is misleading, not just the process that produced it.
The short answer
The instinct to shrug off cheating when the candidate "still passes" misunderstands what an interview score is supposed to measure. A coding interview is not a pass or fail gate for its own sake, it is a proxy for whether the candidate can independently do the work once hired. If the passing score was earned with real-time AI help, the proxy is broken even though the number on paper looks fine. The company is not hiring the score, it is hiring the person, and the person may not be able to reproduce that performance without the same outside help on day one.
This is the exact pattern Fabric documented: 38.5% of candidates were flagged for AI-cheating behavior across 19,368 interviews between July 2025 and January 2026, and 48% in software engineering specifically. Of those flagged software engineering candidates, 61% still scored above the passing threshold. That means the scoring system, used alone, systematically misses more than half of the fraud that proctoring evidence catches.
The evidence
- Fabric (19,368 interviews, Jul 2025 to Jan 2026): 38.5% of all candidates flagged for AI-cheating behavior; 48% in software engineering.
- 61% of flagged software engineering candidates still scored above the passing threshold, meaning the score alone would have led to an offer.
- Karat: 80% of candidates use LLMs during coding tests that explicitly ban them, suggesting the passing-score problem is widespread, not an edge case.
- CodeSignal: technical assessment cheating rate doubled year over year, from 16% to 35%, meaning the share of inflated passing scores is growing, not shrinking.
Score alone vs. score plus proctoring evidence
| Signal used | What it tells you | Risk of a false "pass" |
|---|---|---|
| Coding test score alone | Whether the submitted solution works | High, per Fabric's 61% figure |
| Coding test score + resume | Whether the solution works and looks consistent with experience | Still high, resumes can be embellished too |
| Coding test score + proctoring evidence | Whether the solution works and whether the candidate produced it independently | Low, behavior signals catch what the score misses |
Step-by-step: acting on a passing score with a fraud flag
- Do not treat a passing score as a final answer. Check whether a behavior or identity flag exists alongside it before moving the candidate forward.
- Re-interview live, unscripted, if a flag exists. Ask the candidate to explain their own solution in a different context; genuine understanding survives this, memorized or assisted answers usually do not.
- Weight the trust report alongside the score in the hiring decision, not as a tiebreaker used only when scores are close.
- Document the decision either way. If the company proceeds despite a flag, record why, in case the hire underperforms later and the decision is reviewed.
- Track the flag rate over time. A rising trend across roles, like the CodeSignal 16% to 35% jump, is a signal to strengthen the process, not tighten one-off cases.
FAQ
Does a passing score with a flag always mean the candidate lied? Not always. It means the score cannot be trusted as clean evidence on its own, which is different from proof of intentional fraud. The next step is a live, unscripted follow-up, not an automatic rejection.
Why would a company hire someone despite a fraud flag? Sometimes teams weigh other strong signals, like reference checks or a strong live follow-up conversation, and decide the risk is acceptable. The mistake is not weighing it at all.
Is this specific to software engineering interviews? Software engineering shows the highest flag rate at 48% in the Fabric data, but the same passing-score-hides-cheating dynamic applies to any interview using LLM-generated answers.
How much does a flagged-but-hired candidate typically cost later? This varies by role and cannot be stated as a single number without invented data, but the underlying risk is a hire who cannot independently reproduce the skill the score suggested they had.
What is the fastest way to catch this before an offer goes out? Run identity and behavior proctoring during the same session as the technical score, so both signals are available together before any hiring decision is made.