Помощь ИИ на технических собеседованиях: как компании ее выявляют и предотвращают

A candidate completes the live coding task without a mistake, answers the system design questions confidently and receives a strong recommendation from the panel. A month after joining the team, the same person has trouble with tasks that should be routine for someone at their level.

Situations like this have always happened from time to time, and usually they come down to a poor hiring decision. Over the last two years, however, another explanation has become common. AI assistants can now listen to interview questions, read the screen and suggest answers in real time, and some of them are designed so that the interviewer cannot see them during a video call.

Below we look at how widespread AI-assisted cheating is, what it looks like in practice, which detection methods are useful and how companies are changing their technical interviews so that a hidden assistant gives the candidate little advantage.

How common is it?

There is still not much reliable data, and some of the most detailed figures come from companies that sell detection software, so they should be read with some caution. An analysis by FabricHQ of 19,368 interviews found signs of cheating in 38.5% of them, and the rate tripled between July and September 2025.

According to the same analysis, technical interviews are affected far more than others: the cheating rate was 48% for technical roles and 12% for sales roles. This is not surprising, because technical interviews often use standard questions with clear correct answers, and AI tools handle those very well. Candidates with up to five years of experience were flagged almost twice as often as senior specialists.

Independent research shows a similar trend. Gartner expects that by 2028 one in four candidate profiles worldwide may be fake. In its survey of 3,000 job seekers, 6% admitted to some form of interview fraud, for example having another person attend the interview in their place, HR Dive reports. About 40% said they had used AI at some stage of their application.

What AI-assisted cheating looks like now

A few years ago, cheating in a remote interview usually meant a second monitor or a quick search in another browser tab. The tools available today are considerably more advanced. FabricHQ’s data gives an approximate breakdown of the methods it detected:

Proxy candidates are a separate and more serious problem. In these cases another person takes the interview, or a deepfake is used to hide the real candidate’s face. This happens much less often, but when it does, the company may end up hiring someone it has never actually interviewed.

Why this matters for employers

One figure from the FabricHQ analysis stands out: 61% of the candidates flagged for cheating scored above the passing threshold. A standard interview process would have moved most of them to the next stage, which means the interview was partly assessing how well the candidate used the assistant rather than what they know.

A wrong technical hire costs a lot. The company pays for onboarding, salary and a repeated search. Senior engineers lose time reviewing and fixing the new colleague’s work, and the team slows down. Security is a concern as well, since new employees usually get access to repositories, client systems and internal tools within their first days.

For companies hiring in Belarus, the risk is particularly relevant because many technical interviews are held online as standard. Clients who work through IT outstaffing often do not meet specialists in person before they start, so the assessment process has to be reliable on its own.

Warning signs interviewers notice

Any of the following signs can have an innocent explanation, so they should prompt further questions rather than a conclusion:

• The candidate’s eyes move steadily across one area of the screen before each answer, as though they are reading.
• Every answer, whether the question is simple or complex, begins after the same short pause.
• The first answer sounds polished, but the candidate cannot go further when asked why they chose that approach or what they would change.
• The candidate has difficulty explaining decisions in projects listed on their CV.
• A standard problem is solved perfectly, but a small change to the requirements leaves the candidate unsure what to do.

Interviewers should also keep in mind that anxiety, speaking in a second language or neurodivergence can lead to long pauses and unusual eye contact. Experienced interviewers use these observations to change the direction of the conversation and only draw conclusions once they have more information.

Detection tools and their limits

There are now many products that claim to detect AI-assisted cheating. Most fall into one of these groups:

• Proctoring and browser lockdown tools, which limit what the candidate can open during an online test.
• Code and text analysis, which flags answers that appear AI-generated or closely match known solutions.
• Behavioural analytics, which look at eye movement, typing patterns and how quickly the candidate responds.
• Identity verification, which confirms that the person in the interview is the person named in the documents.

Of these, identity verification is the least used. Gartner data cited by Greenhouse shows that only 7% of recruiting leaders used identity document validation software in 2024, while 54% agreed that AI had made candidate fraud a much bigger problem.

Detection tools have clear limitations. Overlay assistants are built to avoid them, automated flags sometimes mark honest candidates as suspicious, and intensive monitoring raises privacy concerns that candidates are noticing more and more. These tools are useful as an additional input for the interviewer, but the final judgement still has to come from people.

Changing the technical interview

In our experience, changing the interview itself does more to reduce cheating than adding monitoring. A question that can be answered with a single prompt is easy to outsource to an assistant, while questions that require context, judgement and reasoning out loud are much harder to fake. If you are rebuilding your process, HR consulting can help you design interview stages and scorecards around the changes below.

1. Discuss the candidate’s own work

Set aside part of the interview for a past project, a technical decision the candidate made or code they wrote. Ask why they chose a particular approach and what they would do differently today. A person who did the work can discuss it in detail and answer follow-up questions, which is very difficult to do with an assistant.

2. Change the requirements during the task

When the candidate has a working solution, add a new constraint, introduce a bug or ask them to extend the code. Debugging and adapting existing code shows how well someone understands it much more clearly than writing a familiar algorithm from memory.

3. Use tasks based on your own work

Classic algorithm problems are exactly the kind of material AI handles best. Tasks built around your product, your industry or a typical situation in your team are harder to pass with outside help, and they also show more accurately how the candidate would perform in the job.

4. Turn the task into a conversation

In pair programming, the interviewer and the candidate work on the problem together and discuss each step. This format makes hidden assistance hard to use, and the way the candidate explains their thinking is as informative as the code they produce.

5. Adapt the assessment to the role

Each role needs its own approach. With QA automation engineers, asking the candidate to review an existing test framework and point out its weaknesses works well. For leadership roles, a discussion of real trade-offs and team situations tells you more than coding speed.

6. Verify identity and add a final meeting

Check the candidate’s identity at the beginning of the process and once more before making an offer. For senior or sensitive positions, consider a final meeting in person or on camera with the future manager. When hiring a tech lead, this conversation is also a good opportunity to understand how the candidate approaches architecture and managing people.

Setting a clear AI policy for candidates

Some companies have chosen to allow AI in interviews instead of banning it. Canva, for instance, replaced its traditional computer science fundamentals interview with a format in which candidates use their preferred AI tools to solve realistic product problems, its engineering team explains. Interviewers then assess how candidates handle vague requirements, evaluate the AI’s suggestions and correct mistakes in the generated code.

Whichever approach you take, candidates should know the rules. A fair policy usually covers three points:

• Candidates are told in advance whether AI is allowed, preferably in the interview invitation.
• The policy makes clear what counts as normal preparation and what counts as help during the interview.
• Every candidate is assessed under the same rules, and the reasons for each decision are recorded.

Engineers use AI tools in their daily work, and that is unlikely to change. For most teams, the more useful question is whether a candidate understands the code an AI tool produces and is able to take responsibility for it.

How a recruitment partner can help

An agency that specialises in HTP recruitment knows what candidates at a given level in a given role can usually do, so exaggerated profiles are easier to recognise at an early stage.

A number of problems can be identified before the technical interview. Structured screening calls, reference checks with former managers and confirmation of previous employment make it much harder to build a false profile or to rely entirely on an assistant.

Market data can also help. If a candidate’s experience, skills and salary expectations do not match each other, it is reasonable to look more closely. Our developer salary benchmarks for 2026 can serve as a reference point for this comparison.

Summary

AI-assisted cheating has become a real issue in technical hiring, particularly in remote interviews and for junior positions. Detection tools are useful, but the assistants candidates use are improving quickly too, so relying on detection alone is unlikely to be enough.

Companies get better results when they change the interview itself: they talk about the candidate’s real experience, change tasks as they go, use problems from their own work and explain their AI policy openly. These changes reduce cheating and, as a side effect, give a more accurate idea of how a candidate will perform once hired.

Looking for candidates you can rely on?

We screen and verify technical candidates before they reach your interviews, so your team spends its time on people whose skills have already been checked. Contact our team to discuss your next hire.