One in Three Hiring Managers Has Interviewed a Suspected Deepfake. Here's What That Costs Real Candidates.

In July 2024, KnowBe4 — a cybersecurity firm that trains organizations to recognize threats — discovered it had hired a North Korean operative as a software engineer. The person passed a standard background check. They provided references. They cleared four separate video interviews. Nothing in the hiring process raised a flag. What surfaced the fraud was endpoint detection software that flagged malware being loaded onto the company's newly shipped laptop within hours of its arrival.

KnowBe4 published the incident as a warning. The DOJ has since prosecuted facilitators behind larger-scale versions of the same scheme: North Korean IT workers placed inside more than 100 U.S. companies using stolen American identities, generating more than $5 million for the regime. Two U.S. nationals were sentenced to multi-year prison terms for running laptop farms — residential addresses where company hardware was received and remote desktop software was installed so overseas workers could appear to be operating locally.

Those prosecuted cases are a slice of a broader pattern. A 2026 Greenhouse survey of more than 4,000 hiring professionals found that 31 percent have personally interviewed someone they suspected or confirmed was using deepfake technology. Ninety-one percent reported encountering or suspecting AI-generated answers during online video interviews. Misconduct signals in online hiring have grown more than 30 percent year-over-year.

The fraud is real and widespread. Companies have responded with structural changes. And those changes affect every candidate running a remote job search — including the ones who are exactly who they say they are.

Why Deepfakes Are Effective Enough to Be the Problem

The mechanics are less exotic than they sound. AI-enhanced video, voice cloning, and real-time face-swapping tools have become accessible to anyone willing to use them. Research published by Corsound AI documented that someone with no prior image-manipulation experience can assemble a fake video candidate capable of passing a remote interview in approximately 70 minutes. Detection remains poor: a Wiley meta-analysis of deepfake identification studies found that untrained humans correctly identify deepfakes only 55 percent of the time — barely better than chance. Hiring managers going on gut instinct during a 30-minute Zoom call are operating at the edge of that window.

For identity fraud that doesn't rely on deepfakes at all — proxy interviews, where one person interviews on behalf of another — detection is harder still. Sixty-two percent of hiring professionals now say candidates are better at faking than recruiters are at detecting it, according to research from Sherlock AI.

The North Korean scheme worked partly because none of the individual indicators were alarming in isolation: a real U.S. identity, video interviews routed through a domestic connection, and a working history that was real — it just belonged to someone else. KnowBe4's process would have caught many fraud attempts. It didn't catch this one.

What Companies Are Changing

The industry response has been structural, not just technical.

Mandatory in-person rounds have returned. Google reintroduced at least one in-person interview for all candidates in 2026; CEO Sundar Pichai confirmed it explicitly. McKinsey has required an in-person meeting before any offer for roughly 18 months. The Greenhouse data found that 72 percent of recruiting leaders are now conducting in-person interviews specifically as an anti-fraud measure. An in-person round can't be proxied by someone running a remote setup from abroad — it's a blunt instrument, and that's its value.

Identity verification has entered the interview process. Enterprise platforms including HireVue and Spark Hire now offer live ID verification features: candidates appear on camera, hold a government-issued document, and the platform runs a facial match before questions begin. Some companies run this as a default step for all remote first-round screens; others apply it only to technical or sensitive roles.

Work-sample assessments in constrained environments have expanded. For engineering and analytical roles, AI assistance during a conversational interview is difficult to prevent at scale. A monitored, in-browser coding environment is easier to secure. Companies that previously evaluated technical candidates via take-home projects now run those assessments live — video on, with platforms that detect window-switching or pasted code blocks from external sources.

Processes have lengthened. Part of the industry's caution shows up in timelines. More rounds, more human touchpoints, and more verification steps each add days. For candidates, that means more total hours invested per process and more points where inconsistency can surface.

What This Costs Legitimate Candidates

None of these changes were designed to make the experience harder for honest candidates. All of them do.

In-person requirements add travel costs and logistics that candidates absorb. Someone applying for a remote role may be required to fly in for a final round before any offer exists. Candidates without geographic flexibility near major employment centers find that more processes have become inaccessible. The time investment per application cycle has increased while the conversion math has stayed difficult.

Identity verification steps that would have felt unusual two years ago now appear as routine process steps. If you've received a request to verify your identity before a screening call and wondered what you did to raise suspicion, the answer is usually nothing — the same step is going to every candidate.

The subtler cost falls on depth. Hiring managers calibrated to detect inconsistency are scrutinizing answer quality more carefully than before the fraud wave. A candidate whose responses seem unusually polished in round one, or who struggles under follow-up on answers they delivered compellingly a week earlier, now raises a flag that the same candidate would not have raised two years ago. The system is tuned to find gaps: between the person on screen and the person in a room, between the first-round answer and the fourth-round follow-up.

For legitimate candidates, the consistent depth the new environment rewards is exactly what genuine preparation produces. The fraud problem has accidentally created conditions that separate candidates who know their material from candidates who borrowed it.

How to Compete Honestly in a Fraud-Suspicious Market

The adjustments are practical.

Treat in-person as likely, not exceptional. If you're targeting any role at a major employer, plan for at least one in-person round — even if the job is listed as fully remote. Build your preparation assuming you'll be in a room with no notes, no browser tabs, and multiple people evaluating whether the person who arrived matches the one who interviewed on video.

Know each company's AI policy before the call. Amazon requires an explicit acknowledgment that no AI tools are in use and disqualifies candidates who violate the policy. Goldman Sachs bans AI assistance. Meta explicitly permits it. Anthropic allows AI for preparation but not for real-time use during interviews. These policies vary enough that assuming anything is a risk. Checking in advance removes an uncertainty that doesn't need to exist.

Prepare for identity verification as a routine step. Having government ID available before a first-round video call is no longer unusual. Some platforms require you to complete verification before entering the interview environment at all. Treating this as administrative rather than accusatory keeps the process moving.

Build answers you can extend under follow-up. The detection signal that behavioral analysis and attentive human judgment are both calibrated to find is the gap: an answer that's too complete, too polished, or inconsistent in register with how the rest of the conversation has gone. Candidates who prepare real examples — specific, defensible, with concrete numbers and context they can extend when pressed — don't produce that gap. Their follow-up answers are deeper versions of the first ones, not a collapse.

Loading your actual preparation into a structured briefing before each interview — the company's recent news, the role's specific requirements, your specific examples organized for recall — is what makes that consistency achievable under pressure. Meeting Copilot's interview assistant is built around that sequence: a briefing pack assembled from your real material, accessible when it matters, so your own preparation surfaces instead of sitting buried in a closed tab.

The Stable Ground Underneath the Noise

The hiring environment of 2026 runs more identity verification, more in-person rounds, and more scrutiny than the one from two years ago. None of that changes the underlying question every interview is attempting to answer: are you capable of doing this job, and would working with you be good?

The fraud-checking infrastructure filters for people who aren't who they claim to be. It works poorly as an obstacle if you're actually the person you claim to be, doing the work you claim to have done, capable of the skills you claim to possess. Your preparation still determines the outcome — it just has to hold up in a room, not just on a screen.

---

Sources: KnowBe4: How a North Korean Fake IT Worker Tried to Infiltrate Us · DOJ: Two U.S. Nationals Sentenced for Facilitating DPRK IT Worker Schemes · Greenhouse: An AI Trust Crisis in Hiring · Sherlock AI: Rise of AI Interview Fraud in 2026 · Corsound AI: Deepfake Candidate — Data Breach Starts at Interview