Leaked interview questions are exact, or nearly exact, copies of a company’s real assessment or interview questions that have been collected, compiled, and shared, often for a price, so that later candidates can memorize the answer before they ever see the question live. This is a test bank leak: a candidate is not solving a problem in real time, they are recalling a rehearsed answer to a question they already knew was coming. It is a different failure mode than a candidate using ChatGPT mid interview or fabricating a resume, and it needs a different fix.

What Counts as a Test Bank Leak?

A test bank leak happens whenever real assessment content, whether that is a coding problem, a system design prompt, a behavioral interview question, or a full scored rubric, ends up outside the organization that owns it and in front of future candidates before their own interview or test. The leak does not have to be a dramatic data breach. Most of it is candidates simply writing down what they were asked immediately after their interview and posting it somewhere other candidates will find it.

Three things distinguish this from other candidate side dishonesty covered elsewhere on this site:

• It happens before the interview or assessment, not during it, which is what separates it from real time AI assistance like the kind covered in glider.ai’s research on ChatGPT interview cheating.
• It targets the question bank itself, the actual intellectual property behind an assessment program, not the candidate’s identity or resume content, the earlier stage problem covered in glider.ai’s AI resume fraud research.
• It rewards memorization over ability, so a candidate can pass a specific company’s process while having a materially weaker real skill level than their score suggests.

How Do Candidates Get Interview Questions and Answers in Advance?

Candidates get interview questions in advance through a small number of well established channels, ranging from informal community goodwill to fully commercial paid products.

Community forums and country specific boards. Sites like Blind and 1point3acres have run candidate reported interview question threads for years, organized by company and sometimes by role and interview round. A 2025 Blind thread titled “is using leaked interview questions from gl1p3a cheating” shows candidates openly debating whether this practice crosses a line, which suggests it is common enough to need a name and a norm, not a rare edge case.

Dedicated leak aggregator sites. A newer category of site exists specifically to compile and sell this content. LeakCode describes itself as pulling from over ten sources, including paywalled competitors like 1point3acres and LeetCode Premium, and advertises a database of more than 26,900 entries across more than 929 companies for about 100 dollars a year. Gotham Loop takes a more aggressive sourcing approach, claiming to run what it calls exit interviews with recent hires and reporting that it spends roughly 15,000 dollars a month on gift cards to keep its question feed current across more than 160 companies, sold on a 199 dollar a month or 329 dollar lifetime plan. PracHub positions itself as a free alternative to paid options like DarkInterview, crowdsourcing real candidate submitted questions instead of charging for access.

Public code repositories and study guides. Curated lists of company specific interview questions circulate widely on GitHub and in study guide repositories, often assembled from the same underlying candidate reports as the paid sites, just without the paywall or the frequency ranking.

Paid tutors and prep communities. Beyond public sites, some interview prep tutors and private study groups sell access to a running, updated list of exactly what a specific employer has recently asked, functioning as an informal, harder to detect question bank leak.

The common thread across all four channels is the same: a real candidate remembers or records a real question, and that question becomes reusable intelligence for everyone who interviews after them.

How Big Is This Problem, Really?

There is no single, agreed upon industry statistic for what share of candidates use leaked questions, and any exact percentage claimed without a named source should be treated skeptically. What is verifiable is the scale of the supply side. A single leak aggregator citing tens of thousands of entries across hundreds of named companies, and a competing product built around a five figure monthly sourcing budget, both point to real, sustained demand rather than a fringe habit. The existence of a competitive market, with free and paid tiers, frequency rankings, and translated content pulled from international sources, is itself the strongest evidence that test bank leaks are a persistent operational problem for any company running a high volume assessment or interview process, not an occasional embarrassment.

This is also not a new phenomenon in testing more broadly. In psychometrics, the pool of questions an assessment draws from is called an item bank, and test security researchers have long studied what happens when items get exposed. Assess.com describes this as pre knowledge cheating: a candidate studies stolen or leaked exam content ahead of time, often sourced from what the industry calls brain dump sites, where prior test takers post everything they remember. The parallel to hiring assessments is direct. A leaked interview question bank is functionally a brain dump site aimed at a specific employer’s process instead of a certification exam.

How Is This Different From Live AI Assisted Cheating?

A test bank leak is a knowledge problem before the interview. Live AI assisted cheating, the subject of glider.ai’s separate research on ChatGPT interview cheating, is a generation problem during the interview. A candidate who memorized a leaked answer walks in already knowing what is coming and simply recites a rehearsed response. A candidate using ChatGPT in real time is generating an answer on the spot to a question they had never seen before.

Both problems produce the same outcome, a candidate who scores well without demonstrating their real ability, but they need different countermeasures. Detecting live AI use relies on behavioral signals during the session, like unnatural pauses or reading style eye movement. Detecting a memorized answer to a leaked question relies on analyzing the question bank itself and how a candidate’s performance compares against it.

How Can Employers Detect That a Candidate Used Leaked Questions?

Detecting pre knowledge is a pattern problem, not a single tell, and most of the forensic methods come out of large scale certification testing rather than hiring specifically. The core techniques translate well:

• Time score analysis. A candidate who finishes unusually quickly while still scoring at or near the top of the distribution is a classic pre knowledge signal, since genuine problem solving on a question of real difficulty takes measurable time.
• Response matching. Comparing a candidate’s exact wording, code structure, or answer sequence against known leaked content or against other candidates’ submissions can surface suspiciously identical responses.
• Item performance gaps. A candidate who scores perfectly on your most commonly used, most likely to be leaked questions but performs noticeably worse on newly rotated items is showing exactly the gap pre knowledge produces.
• Person fit analysis. When a candidate nails a genuinely hard question but stumbles on an easier one, the pattern does not fit a coherent skill level, which is itself a flag worth reviewing.

None of these signals alone proves a candidate used a leaked question bank. Together, especially layered with the kind of proctoring data covered in glider.ai’s comparison of AI proctoring and human proctoring, they build a reviewable case rather than a guess.

How Do You Rotate and Randomize a Question Bank to Stop Leaks?

The most durable fix is structural: make any single leaked question worth less. That comes down to a handful of concrete practices.

• Build a large enough item bank. A question bank with only a dozen live questions is fully exposed the moment a handful of candidates share their experience. A properly sized item bank, the kind referenced in psychometric literature on item banking, spreads risk across hundreds of interchangeable items so no single leak compromises the whole assessment.
• Rotate and retire exposed items on a schedule. Any question that shows up on a leak aggregator or a public forum thread should be retired from active rotation, not left in circulation indefinitely. Glider AI’s own technical skill test library is built around exactly this kind of large, standardized question pool that supports rotation at scale rather than reusing a fixed, static set.
• Randomize question selection and order per candidate. Serving each candidate a different subset and sequence, drawn from a shared bank, sharply reduces how useful any one candidate’s leaked report is to the next applicant.
• Move from static questions to simulated tasks. A memorized answer to “reverse a linked list” is far more useful to a cheating candidate than a memorized walkthrough of a live coding simulation that reconfigures the underlying project each time. Replacing canned questions with project style coding simulations removes most of the value a leaked answer key would otherwise provide, since the candidate still has to actually operate inside a working environment rather than recite a solution.
• Weight structured, live follow up questioning. A candidate who memorized an answer usually cannot extend or defend it under a slightly different framing asked live, which is the same weakness that also helps catch AI generated answers during an interview.

What Should You Do When You Suspect a Candidate Used Leaked Questions?

Treat it the same way you would treat any other pre hire integrity flag: with a documented, consistent process rather than an ad hoc call.

• Cross check the candidate’s specific questions or code against known leak sites and forum threads when the pattern looks suspicious, rather than assuming coincidence.
• Ask a live, specific follow up that requires extending the original answer, not just repeating it.
• Review timing and scoring data alongside the interviewer’s own notes before making a decision, since a fast, high scoring session on your most exposed questions is a flag worth a second look on its own.
• Retire the specific question from your bank immediately once you know it has been publicly shared, regardless of what you decide about the individual candidate.
• Keep this checkpoint distinct from your identity verification step. A leaked question bank is a content integrity problem, while confirming the person on the call is who they claim to be, the job ID Verify by Glider AI is built for, is a separate and equally necessary checkpoint earlier in the process.

FAQs