Remote hiring removed the natural safeguard
For most of hiring history, assessment integrity was enforced by proximity. When a candidate completes a test in your office, the conditions are visible: one screen, no phone, a room with other people. Not a perfect system — cheating has always been possible — but the friction was real. Effort was required, and effort left traces.
Remote hiring, now standard practice across most industries and nearly universal for technical roles, dissolved that friction. The candidate completes the assessment on their own device, in their own space, at a time they choose. Whether or not anything else is happening on that screen is, by default, unknown to the hiring team.
For years the practical risk was modest. Looking up an answer on Google or asking a colleague was possible but limited — it required context-switching, it added time, and it produced answers the candidate still had to interpret and restate. The signal wasn't clean, but it wasn't catastrophically broken either.
The arrival of capable AI assistants changed the calculus entirely. Pasting an assessment question into a ChatGPT window and receiving a polished, correct answer takes seconds. The candidate doesn't interpret, restate, or even fully read it. They copy. The answer on their screen looks exactly like genuine work. There is no time penalty, no friction, and nothing visible to a remote evaluator. The test measures the AI. The candidate receives the score.
Not all positions are equally vulnerable
Before treating this as a universal crisis, it is worth making a distinction that most coverage of AI cheating misses: the problem is not uniform across role types. Its severity depends entirely on what the assessment is trying to measure — and that varies significantly with seniority.
Senior and architecture roles: measuring thinking, not answers
For senior engineers, architects, technical leads, and similar positions, the most valuable thing to evaluate is not what the candidate knows — it is how they think. What trade-offs do they consider when designing a system? How do they approach an ambiguous problem? What assumptions do they surface? How do they reason about failure modes?
These questions do not have a correct answer to look up. The assessment is the conversation. An AI model can produce a technically competent system design diagram, but it cannot simulate the candidate's reasoning process in real time, explain the judgment calls they made, or respond coherently when challenged on the assumptions they started from. Senior-level assessment is inherently more resistant to AI cheating because the format — open discussion, live whiteboarding, structured technical dialogue — measures a process that cannot be outsourced.
This does not mean senior assessments are immune. A candidate who uses AI to prepare a memorized briefing for an architecture discussion is doing something different from a candidate who can actually reason through that discussion in real time, and experienced interviewers can usually detect the difference. But the core validity of the assessment is much harder to break at this level.
Junior and mid-level roles: this is where it breaks
Junior and mid-level assessments exist for a fundamentally different reason: to verify that a candidate has real working knowledge of the tools, concepts, and methods the role requires. Can they actually write a SQL query that joins three tables correctly? Do they understand what a closure is in JavaScript? Can they implement a basic REST endpoint without looking it up?
These are knowledge questions. They have right and wrong answers. And they are exactly the type of question that AI models answer perfectly, instantly, and with no indication of whether the candidate understood a word of the response.
When a junior candidate uses ChatGPT to complete a technical assessment, the hiring team does not receive a measurement of that candidate's knowledge. They receive a measurement of whether ChatGPT can pass the test — which it can, reliably and at scale. The hiring decision is then made on corrupted data. The company hires someone whose actual knowledge level is unknown. The honest candidates who completed the assessment without assistance are competing against AI output on a playing field that is anything but level.
This is not a theoretical concern. It is happening across the industry, and candidates who use AI assistance have a structural advantage over candidates who choose not to. That is the integrity problem.
What fair play monitoring actually does
The instinctive response to this problem — and one worth resisting — is to treat it as a surveillance challenge: watch everything, catch cheaters, punish them. That framing produces systems that feel adversarial and generate as many false positives as genuine detections.
A more useful framing is data quality. The purpose of an online assessment is to produce reliable measurement data about a candidate. Monitoring exists to flag sessions where the data quality is likely compromised — not to catch bad actors, but to give the HR team information they need to make a sound decision.
In practice, fair play monitoring captures behavioral signals that correlate with off-screen assistance: tab switching, window focus loss, clipboard paste events, unusual timing relative to question complexity, and camera monitoring that confirms physical context. None of these signals is conclusive on its own — a brief focus loss might be a notification, a paste event might be the candidate pasting their own notes. What they produce is a session-level integrity signal that surfaces sessions worth reviewing before the results are acted on.
The outcome is not a binary verdict. It is a flag that says: this session showed patterns consistent with external assistance — review before deciding. For the large majority of sessions, where candidates completed the assessment genuinely, no flag is raised and results are used directly. For flagged sessions, the HR team has the context to schedule a follow-up conversation, ask the candidate to demonstrate their knowledge live, or exercise their judgment about the result.
Fair play protects the honest majority
The conversation around assessment integrity tends to center on catching cheaters. That framing, while understandable, misses the more important party: the majority of candidates who complete assessments honestly.
When a hiring pool includes a significant proportion of AI-assisted results alongside genuine ones, honest candidates are at a structural disadvantage. Their real knowledge level — which may be excellent — competes directly against AI output. If the company uses score rankings to shortlist candidates, the top of the list skews toward AI users regardless of underlying ability. The honest candidate who was outranked by an AI-assisted score never appears in the hiring decision at all.
The company pays for this too, but less immediately. The first feedback arrives months later, when onboarding reveals a knowledge gap that the assessment was supposed to catch. By then the hiring cost is sunk, the team is affected, and the root cause is invisible in the data.
Fair play monitoring is, in this sense, less about catching cheaters and more about preserving the signal the assessment was designed to produce — so that the score a candidate earns actually represents the candidate, not a tool they happened to have open in another tab.
Applying this in your remote hiring process
The practical implication of this analysis is a calibration in how you design assessments and interpret results based on role level.
For senior and architecture roles, lean into formats that are inherently difficult to fake: live technical discussions, architecture review sessions, real-time problem decomposition. These formats are already common at senior levels in many organizations. Their resistance to AI assistance is a structural advantage worth preserving deliberately — not replacing with a take-home test that a model can complete unattended.
For junior and mid-level roles, pair knowledge-assessment formats with fair play monitoring that produces session-level integrity data. Establish a clear internal policy before you need it: flagged sessions get a follow-up question, not an automatic rejection. This protects honest candidates from false positives and gives the hiring team a reliable basis for acting on results.
The combination — assessment format matched to what the role actually requires, paired with monitoring appropriate to that format's integrity requirements — produces reliable hiring data at both levels without turning the process into an adversarial exercise for the candidates who come in good faith.
Platforms like Calibers.ai include real-time proctoring as a built-in layer of remote technical assessments, generating session integrity reports alongside candidate results so HR teams have both the measurement and the quality context in a single workflow.