Can interviewers detect AI copilots? What they see

What counts as an AI copilot in a job interview?

An AI copilot is software that listens to or reads the interview and suggests answers to the candidate while the call is running. CNBC reported in March 2025 that these tools read written and spoken questions and generate code and explanations within seconds. The candidate then repeats or types what the tool produced.

The vendors sell secrecy as the main feature. On October 22, 2026, the homepage of Interview Coder, one of the tools CNBC named, called itself "The No. 1 Undetectable AI For Interviews" and promised "100% undetectability". That is the vendor's own claim, and we found no independent test that confirms or disproves it.

So the useful question is narrower than "can they tell". It's which signals interviewers and hiring platforms actually use, what each one is based on, and how often each one gets an honest candidate wrong. The table below answers that with a published source for every row.

What can interviewers actually detect?

Interviewers work from behavior cues, platform flags and follow-up questions. Each one has a published source, and each has a weakness. Behavior cues are guesses that also fit nervous, honest people, and platform flags are reviewed by a person before anyone decides. Follow-up questions test understanding directly, so they hold up best.

Why is eye movement such a weak signal?

Because honest candidates look away too. A 2005 study in Memory and Cognition by Doherty-Sneddon and Phelps starts from the finding that adults and children avert their gaze when asked questions, "especially when questions are difficult." Hard interview questions are exactly when an honest candidate glances away to think.

That makes a wandering eye a poor basis for a decision on its own. HackerRank seems to agree. Its release notes say that "to reduce false positives," gaze flags are reported as a medium-severity signal so hiring teams can review the activity before deciding. The vendor is telling its own customers not to treat the flag as proof.

One hiring manager told CNBC that eye movement "used to be the biggest giveaway" and that cheating is "harder to detect" now. Read together, the sources point the same way. Watching someone's eyes catches some copilot use, misses some, and flags some people who were only thinking.

What do proctoring tools on coding tests flag?

They flag events and leave the decision to a person. HackerRank's Proctor Mode page lists flags for switching tabs, leaving full screen, connecting a new monitor, a missing or second face on the webcam, and phones or tablets in view. It also says copy and paste is turned off.

The page says the system will "flag suspicious actions in real time," and that suspected outside communication is "flagged for review as potential external assistance." That wording matters. A flag is a prompt for a hiring team to look at a session replay. It isn't a finding. The page we read gives no false-positive rate for any of these checks, so you can't tell from the vendor's own material how often a flag is wrong.

Employers do say they use these systems. Accenture's careers page, checked October 22, 2026, says "We use proctoring tools and AI-detection methods to ensure a level playing field." It also says using AI to complete assessments "unless explicitly stated otherwise" can result in disqualification. The employer AI interview policies page quotes five companies' rules in full.

Why do follow-up questions catch more than software?

A follow-up asks you to explain the answer you just gave, and a copied answer can't explain itself. Brian Ong, Google's vice president of recruiting, said at a February 2025 town hall that interviewers are instructed to probe answers to see whether candidates know what they're talking about, according to an audio recording CNBC reviewed.

The same CNBC report quoted a staffing founder describing candidates whose code "looked OK, but they couldn't describe how they came to the conclusion." No detection software is needed for that. The interviewer asks why, and the candidate either has a reason or doesn't.

This is also how structured interviews are graded. Interviewers write notes against each skill and rate what the answer proved, which the guide to how interview scorecards work explains. An answer you can't defend under one follow-up scores low whether or not anyone suspects a tool.

What happens when an interviewer suspects AI use?

At the companies that have said so publicly, the stated penalty is removal from the process. Amazon's internal guidelines, as reported by Business Insider in February 2025 and quoted by ITPro, warn that "failure to adhere to these guidelines may result in disqualification from the recruitment process." Accenture's page says misuse "can result in disqualification."

There is also a documented case. CNBC reported that after Interview Coder's founder posted a video of himself using the tool in an Amazon interview, Amazon and the other companies that had made him offers rescinded them. The rescinded-offer cases and what happens if you're caught are covered on their own page.

Because cues like eye movement are unreliable, some honest candidates may also be suspected. If an interviewer asks if you're using a tool, answer plainly and offer to share your full screen or walk through your reasoning. A clear explanation of your own work is the best evidence you have.

Are employers allowing AI in interviews instead?

Some are, in specific rounds and with tools the employer provides. Meta told employees in July 2025 that it was building coding interviews where candidates have an AI assistant, according to 404 Media. Business Insider reported in May 2026 that Google will pilot Gemini in a code comprehension round for some software roles.

Meta's internal post, as quoted by 404 Media, said the format "is more representative of the developer environment that our future employees will work in." Business Insider reported that Google's round will assess "AI fluency, including prompt engineering, output validation and debugging." In those rounds, using the provided assistant is the task, and nobody is trying to catch it.

This is where the copilot vendors have a fair point. The people behind two of these tools told CNBC that puzzle-style coding interviews don't reflect how engineers now work, and two of the largest employers have since changed some rounds in that direction.

The point stops at the employer's rules. A tool the interviewer can't see is still against the rules wherever the employer bans it. The list of process changes companies made to counter AI copilots covers the other side.

How can you use AI without any detection risk?

Use it before and after the interview, and use nothing during it that the employer hasn't offered. Practice with AI, research the company, and review your answers once the call is over. None of that happens while you're being assessed, so there's nothing to detect and nothing to explain.

The line is drawn in more detail in where using AI in an interview becomes cheating. The short version is timing. Help that changes what you know before the call is preparation. Help that answers for you during the call is what employers ban.

Prept takes the same line, analysis after the call, never during it, and gives no live help in any form. After an interview, the free 10-minute debrief template helps you write down what was asked and where your answers were thin. If you recorded the call with everyone's consent, Prept's free tier transcribes and debriefs one interview hour (start a free debrief).