A 21-year-old Columbia University student built a business around helping software engineers cheat their way through interviews at Google and Amazon. His tool reads coding questions off the screen, generates the answer, and stays invisible to interview recording software. It charges $60 a month. It was on track to hit $1 million in annual recurring revenue within months of launch.
That’s not a hypothetical. That’s Chungin “Roy” Lee and Interview Coder, reported by CNBC in early 2025.
It’s also the clearest answer to why Google, Cisco, McKinsey, and a growing list of major employers are quietly bringing candidates back into the office.
But here’s the part most coverage of this story leaves out: at the exact same time, Meta, Canva, and even Google itself are doing the opposite for some roles, building AI assistance directly into the interview. Both trends are real. Both are happening right now. This article covers both, with sourcing for every claim.
Quick Answer
- The trigger: AI tools make it trivial to feed candidates real-time answers during remote interviews, while deepfakes and proxy test-takers make identity fraud easier too.
- Gartner’s number: 72.4% of recruiting leaders now run in-person interviews specifically to fight this kind of fraud.
- The companies doing it: Google, Cisco, McKinsey, Amazon, and Deloitte have all added verification steps or in-person rounds since mid-2025.
- The twist: Meta, Canva, and Google are also building AI-assisted rounds into their technical interviews, for different roles and different reasons. This isn’t a simple “AI bad, humans good” story.
- Where it’s heading: roughly a quarter of employers now allow AI in technical interviews, and that share is expected to climb toward half within a couple of years, even as in-person verification rounds also keep growing.
- For recruiters: the practical answer isn’t picking a side. It’s building a hybrid process, AI-assisted screening for scale early on, human or in-person verification where it actually matters.
The Real Trigger: AI Made Virtual Interviews Easy to Fake
Let’s start with the quote that’s been making the rounds.
“Are you interviewing Claude, or are you interviewing the person?”
That’s Nick Bloom, the Stanford economics professor best known for his remote-work research, speaking on the Financial Times’ Economics Show podcast. He was describing something simple: a candidate runs a coding question through an AI chatbot off-screen, then reads the answer back like it’s their own thinking.
Bloom’s point wasn’t just that cheating happens. It’s that remote technical interviews were never built to withstand it.
Here’s why that matters so much right now:
- A Checkr survey of 3,000 hiring managers found 59% suspected candidates had used AI to misrepresent themselves during hiring.
- Fabric, a company that builds AI-interview fraud detection, analyzed 19,368 AI-powered interviews and flagged 38.5% for cheating behavior.
- Only 6% of candidates admit to interview fraud when asked directly, according to Gartner’s own survey of 3,000 job seekers.
Sit with that gap for a second. Detection tools flag more than a third of interviews. Self-reported admission sits at 6%. That gap is basically the whole story in one stat.
And it’s not just coding answers. Gartner is tracking a parallel problem: candidate identity fraud, fake credentials, AI-generated employment histories, and even “proxy” interviews, where one person interviews and a different person shows up for the job. Gartner’s own research predicts that by 2028, one in four job candidate profiles worldwide could be fake in some way.
How Companies Are Actually Catching It
This is the part that rarely makes it past the headline. The verification tactics are getting specific, and a little strange.
Somos, a remote-first telecom and security company, now asks candidates to pan their camera around the room and wave a hand in front of their face mid-interview. It also checks IP addresses, and has caught candidates claiming to be US-based who were actually logging in from abroad. Some candidates have simply ended the call when asked to do this.
Passion.io, a remote-first startup, leans less on physical checks and more on process. Candidates now do live, screen-shared written exercises instead of take-home assignments, after the company found too many take-home submissions were heavily AI-generated.
Nominal goes further still: final interviews happen in person, with engineers writing code on company-provided laptops. No ambiguity about what tool touched the keyboard.
None of these are household names, and that’s sort of the point. This isn’t only a Big Tech story. It’s happening across the board, wherever remote hiring at scale made verification hard.
The Household Names: What Google, Cisco, and McKinsey Are Actually Doing
Now to the part everyone’s already heard a version of, but let’s get the details right, because a lot of what’s circulating online is slightly off.
Here’s the real quote, from Sundar Pichai on the Lex Fridman Podcast in June 2025:
“Given we all work hybrid, I think it’s worth thinking about some fraction of the interviews being in person. We are making sure we’ll introduce at least one round of in-person interviews for people, just to make sure the fundamentals are there.”
Notice what that actually says. Not “we’re banning remote interviews.” Not “AI cheating broke our hiring process.” Just: some fraction of interviews, at least one round. That’s a meaningfully smaller claim than some of the viral social posts summarizing it.
A few more verified details worth knowing:
- The push reportedly started from Google’s own employees. At a February 2026 internal town hall, staff directly asked leadership whether onsite interviews could return.
- Google VP Brian Ong has acknowledged AI created a “fundamental challenge” in recruitment, and that competitors are facing the identical problem.
- There’s a real tradeoff being weighed here, too. By Google’s own internal comparison, virtual interview loops move roughly two weeks faster than in-person ones. Bringing people onsite isn’t free.
Cisco and McKinsey
Both have added in-person rounds specifically to verify technical skill and candidate authenticity, confirmed independently by Computerworld’s August 2025 reporting alongside Google.
Amazon, Anthropic, and Deloitte
- Amazon now requires candidates to formally declare they won’t use unauthorized AI tools during interviews.
- Anthropic, Claude’s own maker, explicitly asks interview candidates not to use AI assistants, a detail that tends to surprise people.
- Deloitte has already rolled out in-person assessments for its UK graduate recruitment program.
The Twist Nobody’s Covering: Some Tech Giants Are Doing the Opposite
Here’s where this story gets genuinely interesting, and where most coverage stops short.
While one part of the industry is pulling interviews back into the office, another part is pushing AI further into the interview room, on purpose.
| Company | What changed | Since |
| Added at least one in-person round, and separately piloted a Gemini-assisted coding round for junior/mid-level roles | Mid-2025 (in-person) / May 2026 (AI-assisted pilot) | |
| Cisco | Added in-person rounds to verify technical skill | 2025 |
| McKinsey | Added in-person rounds to verify technical skill | 2025 |
| Amazon | Requires candidates to declare they won’t use unauthorized AI tools | 2025 |
| Deloitte | In-person assessments for UK graduate program | 2025 |
| Meta | Replaced one of two onsite coding rounds with an AI-assisted round (candidate’s choice of model) | Piloted Oct 2025, expanded through 2026 |
| Canva | Fully replaced its Computer Science Fundamentals round with an AI-Assisted Coding round | June 2025 |
A few things to unpack here:
- Meta didn’t just allow AI, it built it into the interview environment. Candidates now get a real multi-file codebase inside CoderPad with an AI chat window attached, and can choose from several models. At the senior end (E7 and above), the AI-assisted round is the only coding round left.
- Canva says nearly half of its own engineers already use AI daily, so the company redesigned interviews to match how the job actually works. Blind, uncritical acceptance of AI-generated output is treated as a red flag, the opposite failure mode from before.
- Google is doing both at once. An in-person round for verification, and a separate Gemini-assisted pilot for certain roles. That’s not a contradiction, it’s two different tools for two different problems: identity and effort verification on one side, realistic skill assessment on the other.
Zoom out, and the market data backs this up. A 2026 read of the New York job market found a little over a quarter of employers now allow AI in technical interviews, with recruiters expecting that figure to climb toward half within a couple of years.
So the honest framing isn’t “tech giants are going back to in-person.” It’s: tech giants are splitting their interview process into two tracks, one that verifies a human did the thinking, and one that tests how well someone works alongside AI, because both skills now matter on the job.
Where This Is Heading
A few forward-looking data points worth tracking:
- Gartner predicts candidate fraud will keep growing, with as many as 1 in 4 global candidate profiles potentially fake by 2028.
- Gartner’s own HR tech analyst, Emi Chiba, has argued the winning strategy isn’t picking AI or humans, it’s “transparent and consistent communications about the expectations of AI use throughout the recruiting process.”
- Interestingly, candidates themselves don’t seem to mind in-person rounds. Gartner’s data found 62% of candidates say they’re more likely to apply to a role that requires an in-person interview, not less.
Put together, this doesn’t look like a pendulum swinging back to pre-pandemic hiring. It looks like two separate, permanent additions to the hiring process: AI-assisted rounds where collaborating with AI is literally the job, and verification rounds where it isn’t.
What This Means If You’re Hiring Right Now
If you’re a recruiter, HR professional, or hiring manager reading this and wondering what to actually do, the honest answer isn’t “go fully remote” or “go fully in-person.” It’s building a process that uses each format for what it’s actually good at.
A hybrid structure that’s working for a lot of teams right now looks like this:
- Early rounds, scale for volume. Use automated resume screening and remote, AI-assisted video interviews to get through hundreds of applicants efficiently. This is exactly where AI interview platforms built for employers and AI candidate screening tools earn their keep, they’re built for volume, not final judgment calls.
- Middle rounds, test real thinking. Borrow the Meta and Canva playbook even if you’re not a tech giant: give candidates a realistic task, let them use AI the way they would on the job, and watch how they direct it, catch its mistakes, and make decisions. This connects directly to the kind of interview questions you actually ask, which matter more than the format itself.
- Final rounds, verify in person or live. Save in-person or live, unscripted conversation for the decision-stage interview, particularly for senior, high-trust, or high-stress role hires where getting it wrong is expensive.
This maps cleanly onto a well-run full-cycle recruiting process: different verification intensity at different stages, instead of applying the same scrutiny everywhere and slowing the whole funnel down.
One more practical point: whichever approach you choose, Gartner’s advice holds regardless of company size. Tell candidates upfront what AI use is and isn’t allowed, at which stage. Ambiguity is what creates the “was that cheating or was that allowed” arguments in the first place.
Frequently Asked Questions
Which tech companies have gone back to in-person interviews?
Google, Cisco, and McKinsey have all added in-person interview rounds since 2025, specifically to verify technical skill and candidate identity. Amazon now requires candidates to declare they won’t use unauthorized AI tools, and Deloitte has introduced in-person assessments for its UK graduate program. Smaller, remote-first companies like Somos and Nominal have gone further, with some requiring final rounds entirely in person.
Are tech companies banning AI in interviews entirely?
No, and this is the most commonly missed part of the story. Meta and Canva have both built AI assistance directly into parts of their technical interviews, and Google is piloting an AI-assisted coding round alongside its in-person verification round. Most companies are drawing a line between using AI to think for you, which they’re cracking down on, and using AI the way you would on the job, which some are now actively testing for.
How are companies detecting AI use during interviews?
Methods range from simple to elaborate: asking candidates to pan their camera around the room, perform a hand-wave test, share their screen, or complete coding tests on a company-provided laptop. Software tools like BrightHire also monitor signals such as browser-tab switching during remote interviews, though results are typically reviewed by a person rather than used to auto-reject a candidate.
Is using AI in a job interview actually cheating?
It depends entirely on what the employer has disclosed as acceptable. Using AI to generate an answer and presenting it as your own unassisted thinking, when that wasn’t allowed, is widely treated as fraud. Using AI openly, the way Meta’s and Canva’s AI-assisted rounds are designed around, isn’t cheating, it’s the format. The deciding factor is disclosure and consent, not the tool itself.
What is the 80/20 rule in interviewing?
It’s the same Pareto Principle used elsewhere in hiring: roughly 80% of a candidate’s suitability for a role tends to show up through a small set of high-signal moments, like how they reason through an unfamiliar problem, rather than being evenly spread across every interview question. In practice, it’s used to argue for fewer, deeper interview rounds focused on those high-signal moments instead of many shallow ones.
Will in-person interviews become the standard again?
Unlikely, at least not as a full replacement for remote hiring. The data points toward a split process instead: AI-assisted or remote rounds for scale and for roles where AI collaboration is part of the job, and in-person or live verification rounds reserved for final-stage, high-trust decisions. Gartner’s research suggests both trends, rising AI-assisted interviewing and rising in-person verification, are growing at the same time, not replacing one another.