NEW YORK – In retrospect, the interview had gone too well.
The candidate’s answers were polished, if a little scripted. He seemed more than capable for the grant-writing job, so a New York City nonprofit hired him. Within a month, his boss realised he could not do the project he had interviewed for. He froze on basic decisions and was far less nimble than he had seemed from behind a screen.
The manager, who asked not to be identified discussing personnel matters – and because, she says, it is embarrassing to admit you got duped by artificial intelligence (AI) – now feels almost certain the candidate had been covertly consulting a chatbot like ChatGPT during the virtual interviews.
Looking back, the clues were there: Generic answers, language lifted from the organisation’s website and tidy three-part responses to almost every question.
The manager fired him and is on the hunt for his replacement, a time-consuming endeavour. The ordeal also messed with her head. Now every resume looks a little suspicious. Is the applicant talking themselves up or inventing a whole new self? Are they well-prepared or reading a chat-generated script? Is the person behind the screen even real?
Ben Eubanks, chief executive at human resources consultancy Lighthouse Research & Advisory, has heard versions of the same story from multiple employers and hiring managers, who describe candidates who sound brilliant in interviews, only to reveal on the job that the confidence was not theirs alone. In some cases, he says, AI does not fool only the interviewer. It seems to fool the applicant too.
“That sort of thing makes that person feel like, ‘Ah, you know what, I’m sure I can do this,’” he says. Often, they cannot.
This is the new AI-induced hiring paranoia.
Employers have long known that candidates might pad a resume or rehearse their answers. Now the fear goes deeper. Managers worry the entire interaction has been mediated, optimised or even faked by software, leaving managers unsure whether they have evaluated a candidate or merely been shown a convincing simulation of one.
“It’s so much less efficient,” says Rich Braun, a partner at the recruiting firm Delphi-US. “We have to kiss even more frogs to find that prince.”
Almost half of applicants say they are using AI for help at some point in the job search process, according to Gartner data from the end of 2025, up from about 40 per cent in 2024.
Some of that is relatively benign, such as polishing resumes or helping to write cover letters. But 32 per cent are using it to generate text for writing samples and 26 per cent to answer questions on assessments; 13 per cent of respondents admit they are using chatbots in real time during job interviews.
Herval Freire, chief technology officer at the healthcare tech startup Telepatia, calls it teleprompting. The tell, he says, is a candidate’s eyes darting across the screen, and they sound like they are reading.
It became common enough that Freire started throwing in a question about hobbies to see if he can shake applicants out of the trance. When they trip up – and they usually do – “I’m like, ‘Dude, you’re not even listening’,” he says.
AI-enabled deceptions can go further.
There is a growing crop of tools that promise to help would-be engineers cheat on coding tests and technical screens. Charging anywhere from US$10 (S$13) to US$1,800, services such as WhisprGPT and Shadecoder purport to instantly process information – including whatever is on a user’s screen and real-time audio – to solve a variety of coding exercises. Reddit forums give tips on how to evade detection (never switch tabs) and which programmes give the best results (they are all pretty mid, posters say).
For many candidates, the ethical line is not as bright as employers might like.
In a hellish job market, some say they will do anything to get an edge. They will have access to chatbots on the job, they reason, so why shouldn’t they use it to get the job? Plus, hiring is already a game of performance, and isn’t American work culture built on “fake it till you make it”?
But in some cases, AI tools are not enhancing people’s abilities as much as disguising the gaps in them.
Caroline Ogawa, a director at the consulting firm Gartner, says human resources departments have noticed a rise in what are called “regrettable hires”, an effect she attributes in part to people relying too heavily on AI during interviews.
“These are absolutely things that are putting the thumb on the scale for candidates,” she says.
Such errors are not just frustrating, they are expensive. Filling a role typically costs about US$1,300, according to data from the Society for Human Resource Management. This does not account for sunk onboarding costs, weeks of lost productivity and the time bosses spend looking for potential replacements.
To that end, companies are rethinking their approach.
Earlier in 2026, Julia Enthoven realised she had to overhaul the hiring process she had been using for the past seven years at Kapwing, the video editing startup she co-founded and runs in San Francisco.
In the past, candidates would receive a take-home assessment, where they had to solve a coding problem. Pre-AI, about a quarter of candidates both completed and passed the assessment. That helped ensure that only the best potential hires made it to the final round, which typically consisted of two in-person interviews.
In 2025, Enthoven noticed the dynamics around the take-home assessment started to shift. Almost everyone handed in the assignment, and 80 per cent of people passed, securing them an invitation to the office.
That led to an influx of interviews with people who clearly were not qualified. Some of the conversations were so painful, hiring managers cut them short.
“It got to a place where it was a waste of time for both sides,” Enthoven says.
Then Kapwing almost made an offer to someone who Enthoven says was possibly fabricating their identity. After passing the coding test, the candidate, who called himself Dwayne, said he could not travel for an in-person interview because he was caring for a sick relative in Utah.
They met virtually throughout the interview process instead. Alarm bells went off when Enthoven went to call a reference.
First, she looked at the reference’s LinkedIn page and found it had only recently been created and had just a few dozen connections. Then, when she got him on the phone, he could not speak in any detail about the candidate’s work history. At that point, it was too fishy to move forward. Enthoven knew she made the right decision when she sent a rejection note and “Dwayne” responded just saying: “Sounds good.”
Enthoven is hesitant to ban AI use in job interviews outright, because it is become so central to software development.
She has heard of other tech organisations requiring in-person tests, proctored like the SATs, or sending candidates cameras to mount behind them so a company has full visibility into their workstation as they complete the virtual process. Other companies are doing AI-fraud training or buying AI-detection software that tracks keystrokes and eye movements or that allows employers remote access to people’s desktops.
Some employers are drawing hard lines.
Target Corp tells candidates their interview responses must reflect their “personal understanding and judgment, without reliance on external AI-related assistance”. Alphabet chief executive Sundar Pichai said last summer that Google is moving to include at least one face-to-face interview round “to make sure the fundamentals are there”. Almost all software engineers now have at least one in-person interview, the company told Bloomberg Businessweek.
L’Oreal has designated interviews an “AI-free zone” and begun requiring candidates to complete at least one round in person. Perhaps surprisingly, Anthropic bars the use of AI in live interviews and take-home evaluations unless explicitly authorised.
Freire, of Telepatia, now has candidates work through coding challenges live while sharing their screen to assess the person’s ability to communicate their thought process in real time.
Enthoven has not completely ruled out these measures but says she hopes they will not be necessary. For now, the company dropped its take-home test and hired an outside recruiting firm to help with the initial round of vetting.
Alexa Loken, a career coach in San Luis Obispo, California, has been surprised by the intensity of some managers’ anger about the idea of AI assistance.
Loken, who teaches an AI literacy course at a local college, advises students to bring handwritten notes to video interviews to signal they did their own work. That way, their interviewer will be less likely to think they’re doing anything disreputable, including reading answers fed to them by a chatbot.
Mohit Bhende, who co-founded Karat, a company that conducts technical interviews on employers’ behalf, has decided to avoid invasive counter-measures such as tracking eye movement and speech patterns.
“It creates a fundamentally adversarial approach in the interview,” Bhende says.
For job seekers navigating an already miserable market, being presumed a fraud merely adds another indignity. In addition to proving they are right for a job so many others are competing to get, they now also have to prove their polish is not AI-generated. Bloomberg
