Four ways candidates beat online assessments and how to handle each one

7:34 pm 17/Aug/26

Four ways candidates beat online assessments and how to handle each one

AI-assisted cheating in hiring assessments is now a market with paid products and communities. VCO Group maps four common forms and how to handle each.

Cheating in online hiring assessments has moved from individual behaviour to a market with paid products, release cycles and communities sharing techniques for getting past monitoring controls. VCO Group summarises the four most common forms seen in 2026 and the principle for handling them: shift from verifying the answer to verifying the reasoning.

Table of Contents

The scale of the problem has changed, the defences have not

For many years, cheating in online hiring assessments was scattered behaviour with no tooling behind it. The corresponding defences were equally simple: limit the time, shuffle the question order, and trust that most candidates are honest.

The picture in 2026 is entirely different. A category of software products now exists that is purpose-built to assist candidates while they sit an assessment and while they interview, with tiered pricing, feature release cycles, and user communities sharing techniques for getting past proctoring software.

Industry surveys record that around 59 per cent of hiring managers suspect candidates of using artificial intelligence tools to overstate their ability in live assessments. Another study records that around half of job seekers admit to having used generative tools to present their skills differently from reality during an application or an assessment.

For Young Talent Program and Management Trainee schemes, the consequence is not only hiring the wrong person. It is that the entire score dataset for that season loses its reliability, and every comparison between seasons goes with it.

The four most common forms and the mechanism behind each

The four forms below are the most frequently recorded in the industry literature in 2026. They differ in mechanism, so they also require four different responses.

Four ways of beating an online assessment and how to handle each

Form	Mechanism	Effective response
Answers written by a generative tool	The candidate passes the question to a tool and submits the result it returns	Add a short step asking candidates to explain their own choices. Most cases surface within the first two questions.
Someone else sitting the assessment	Another person logs in and completes the assessment on the candidate's behalf	Verify identity at the moment of the assessment and cross-check again at an in-person interview round.
Assistance outside the camera frame	A second device or a person sitting outside the camera's field of view	Design the assessment with a real-time response element, so that looking things up externally becomes slower than the required answering pace.
Virtual machines and hidden environments	Running a second session that the proctoring software cannot see	Do not fight this by detecting specific software; measure anomalies in response pacing and behavioural patterns instead.

The first three forms can be handled through assessment design. The fourth is a technical arms race with no end point, so the response is to reduce the value of cheating rather than trying to detect all of it.

The replacement principle: verify the process, not the answer

What all four forms have in common is that they attack the final answer. The candidate submits a correct answer without having gone through the process that produces it. When an assessment scores only the answer, the only defence left is monitoring, and monitoring is a race in which the defending side is always one step behind.

The more effective direction is to change what is being scored. Instead of asking candidates what the answer is, ask why they chose it, what alternatives they considered, and what they would do if one condition in the question changed.

The mechanism behind this principle is quite specific. Current assistive tools handle a standalone question well, but struggle to hold the thread of context across several consecutive queries about the same situation, particularly when the questions probe trade-offs between options. A candidate who genuinely understands the problem defends their choice comfortably, while a candidate reading back a supplied answer is exposed very quickly.

In practice, a short follow-up step of around ten minutes, placed after the assessment and applied only to the high-scoring band, is enough to filter out most cases without making the process heavier for every candidate.

Three tiers of integrity for a large-scale programme

For a programme receiving several thousand applications, the highest level of control cannot be applied to every candidate. The cost does not allow it, and heavy controls at the first round reduce the completion rate among honest candidates as well.

The sensible allocation is three tiers, with the level of control rising in line with how much the decision matters.

Tier one applies to all candidates at the aptitude testing round and consists of measures that cost close to nothing: state the policy on assistive tools clearly in the assessment instructions, shuffle question and option order, and record response pacing to detect anomalies. Stating the policy up front has a larger effect than many people expect, because it moves the behaviour from a grey area into an area where a warning has been given.

Tier two applies to the group that passes the test round, and consists of identity verification at the point of the next assessment plus a short follow-up step asking candidates to explain their choices. Tier three applies to the group reaching the final round, where the cross-check happens directly in the in-person interview or in the Assessment Centre, and the cost of verification is already built into the process.

One question to answer before designing any measure

Before deploying any measure, there is a question that needs settling at decision-making level: for this role, is a candidate's use of artificial intelligence tools behaviour to be prevented, or an ability to be assessed.

For roles where using assistive tools is part of the daily work, the right question is not whether candidates use them but how they use them. In that case, the sensible design is to permit use openly and to score the quality of that use.

For assessment sections intended to measure underlying ability such as reasoning and verbal processing, the use of assistive tools destroys the meaning of the measurement entirely, so it needs to be clearly prevented.

Most programmes contain both kinds of content within the same process. The policy should therefore be written per assessment section rather than as a single statement covering the whole programme, and it needs to be communicated to candidates in clear language from the initial instructions onwards.

Three things to do

State the policy on assistive tools per assessment section within the assessment instructions themselves, rather than writing one general policy for the whole programme.
Add a short follow-up step of around ten minutes for the high-scoring band, asking candidates to explain their own choices and trade-offs rather than repeat the answer.
Allocate control across three rising tiers, so that the cost of verification is concentrated on the candidate group where the decision matters most.

Frequently asked questions

How do you detect candidates using AI in a hiring assessment?

The most effective approach is not to detect the tool but to change what is being scored. Add a short step asking candidates why they chose that option and what they would do if the conditions in the question changed. Assistive tools struggle to hold the thread of context across several queries about the same situation.

Should candidates be banned from using AI altogether during an assessment?

It depends on the assessment section. For sections measuring underlying ability such as reasoning and verbal processing, the use of assistive tools destroys the meaning of the measurement and should be clearly prevented. For work-simulation sections where using such tools is daily activity in that role, it is more sensible to permit use openly and score the quality of that use.

What level of integrity control should a large-scale programme apply?

Three rising tiers. Tier one covers all candidates at the test round with measures that cost close to nothing. Tier two covers the group that passes, with identity verification and a short follow-up step. Tier three covers the group reaching the final round, where the cross-check happens directly within the existing process.

About VCO Group

VCO Group operates the assessment infrastructure behind Young Talent Programmes and Management Trainee schemes in Vietnam, covering application intake, aptitude testing, automated interviewing and Assessment Centres. VCO Group serves young talent and candidates with under three years of experience, working with large enterprises and multinationals in Vietnam. Its assessment and candidate management platform is called Testalents. Head office in Ho Chi Minh City.

If you would like to benchmark your own selection data against the market, or discuss the Young Talent Program Insight Report, contact Mr. Dung on 0334 889 068 or visit vcogroup.com.vn.