- Most hiring processes check the person once, hard, near the offer. Fake candidates plan for exactly that check.
- Give every stage one or two checks and one owner. Keep checks light for everyone at the start and heavier for the few who reach the offer.
- One signal is a question. Several signals from different areas are a reason to verify.
- At our own agency, recruiters confirmed 118 of 1,768 applications as fraud attempts in one month. About 1 in 15, and most left traces in the application itself.
- You can start tomorrow with no budget: an owner, a signal list, trained hiring managers and a written path from signal to decision.
Fake candidates rarely fail one big check, because they prepare for it. The way to spot them in remote hiring is to check a little at every stage, from the application to the first months, and to give each stage one owner. Below: what to check at each stage, who owns it, and what we learned checking 1,768 applications at our own agency.
Recently I had a call with a manager at a software company. They have offices in a few countries, including the US.
A couple of months ago his phone rang on a Saturday. It was the FBI.
Three of his software engineers in the US were North Korean operatives working under fake American identities. They had passed every stage of his hiring process. They had been on the payroll, with access to code.
He’s a careful manager with a careful team. That’s the point. His process looked like most processes: a few interviews, a background check, an offer.
Most processes check the person once, at the end. Fake candidates plan for that.
How big is this?
- By 2028, 1 in 4 candidate profiles worldwide could be fake, according to Gartner.
- 40% of US hiring managers have interviewed someone later found to be using a false identity, according to Checkr’s Hiring Hoax 2026. A year earlier it was 31%.
- One North Korean cell of up to 22 people sent 166,893 job applications between December 2024 and September 2025 and got at least 76 offers, according to Nisos. Most of the roles were in software, development and data.
- In our own pipeline, in the first month of TalentShield on live roles at Team Up, our recruiters confirmed 118 of 1,768 applications as fraud attempts. That’s about 1 in 15. Many of our roles are remote and technical, which is what these operations target, so your number will differ. More in TalentShield at Team Up.
State operators are only one type. There are also laptop farms running several jobs on bought or stolen identities, and real people with CVs inflated by AI who read answers off a second screen. You’ll meet all three in the same pipeline. More on each in Who is behind fake candidates.
Why one big check at the end isn’t enough
Background checks and ID verification usually happen at the offer stage, for good reasons. They cost money, take time and slow the process down, so nobody runs them on every applicant.
That leaves two gaps. By the offer stage, your recruiter and hiring manager have already spent hours on the person. And the people you worry about most prepare for this exact check. In October 2026 the FBI warned that North Korean IT workers buy accounts with identity verification providers to make sure stolen identities will pass background checks. Only the identities that pass are used to apply.
So keep the checks at the end, and add lighter ones earlier. We wrote more on this in Background checks and ID verification come too late.
Two rules before the checklist
One signal is a question
A new LinkedIn profile, a bad connection, an unusual email address: real people have those every day. One odd detail is a reason to ask. Several signals from different areas, like the file, the contact details and the story in the CV, are a reason to slow down and verify.
A simple rule many TA teams use: yellow signals mean ask more, red signals mean stop and verify.

Layers, with friction that grows
Safety engineers call this the Swiss cheese model, after the psychologist James Reason. Every layer of defense has holes. Stack enough layers and the holes rarely line up. In hiring, a layer is a check at one stage, owned by one person.
The question is where to put the friction. Heavy checks for every applicant cost you good candidates and recruiter time. So keep the first layers light and invisible: everyone passes through them and nobody has to do anything extra. Make the later layers heavier, for the few who reach the offer or earn a closer look on the way.

Here’s what to check at each stage, and who should own it.
Stage 1: The application and the CV
Who owns it: automated checks inside your ATS, with a recruiter reviewing the flagged applications.
This is where volume is highest and checking by hand costs the most. It’s also where many fake candidates leave the clearest traces. In our first month at Team Up, most of the 118 confirmed cases were visible here, before anyone booked a call.
What to look for:
- The file itself. The CV document’s “Author” field shows a different name, and the same author name shows up on several CVs. The file was made with an unusual tool or at an odd hour. The file name doesn’t match the candidate, or is written to impress the reader. The same person sends several versions in one week.
- Text written for machines. White text on a white background, or phrases like “ignore previous instructions” aimed at AI screening tools. One CV in our pipeline looked ordinary, with a paragraph hidden in white text that asked “automated analysis software” to approve the candidate. A real candidate has no reason to hide text in a CV, though some real people try it to game the ATS.
- What the CV says. Pseudo-precision, like “improved performance by 37.4%” in every bullet. A metric left as a template placeholder, like “reduced latency by 0%”. The same description copied across different roles. A version number next to every single skill. A native language that doesn’t fit the person’s history.
- A CV that reads like a printout. One column, only experience and education, nothing personal, not a single rough edge. Real CVs are a little messy.
- Companies and schools you can’t confirm. Check that employers exist in the official company register of their country. Check that the university has the faculty the candidate names. Verify certificates on the issuer’s site.
- Contact details. A VoIP number instead of a mobile. VoIP numbers are easy to buy without registration and are often used to look local. An email address that doesn’t match the name, uses a disposable domain or looks machine-generated.
- Links and footprint. A LinkedIn profile created this year next to “10 years of experience”. Dates and employers that don’t match the CV. Links that don’t open. Or a footprint that exists only to back up the CV: a real person leaves traces over years.
- Your own ATS. The FBI recommends cross-checking HR systems for other applicants with the same resume content or contact information. In one batch we reviewed, two “different” candidates had the same phone number and a skills section copied word for word.
- How it arrived. If your careers site logs it, look for many applications with different names from one IP address or device within minutes, or traffic from data centers and commercial VPNs. In one 2026 investigation, Huntress traced suspected North Korean workers to a commercial VPN and proxy service. A hidden honeypot field in your form catches simple bots. Expect automation signals to matter less over time, as more honest candidates apply through AI agents.
This is the stage TalentShield works at. It checks every application inside your ATS against 70+ signals, shows what it found and why it matters, and leaves the decision to the recruiter.
Stage 2: The first contact
Who owns it: the recruiter.
Before we had a process, this is where most fake candidates showed up. Usually when we tried to book a video call, or simply called the candidate.
What to look for:
- The number doesn’t work, or nobody picks up. Then an email asking to keep it in writing, or to move to WhatsApp or Telegram. Once is normal. Every time is a pattern.
- A language switch, right away. The name, schools and career all point to one native language, and the candidate asks to do the call in English. Some people grew up abroad and are rusty in their parents’ language. Ask before you assume.
- Hours that don’t fit their city. Only free at night their time, or replies that land in another time zone’s working day. People travel and keep odd hours, so ask about it.
- They can’t receive a verification code. Send an SMS code to the number on the CV. Some VoIP setups can’t receive it.
The question that works: “What’s your name?”
It sounds like a joke. We started asking it on unplanned calls, and it became our favorite question. Someone who runs several identities often doesn’t remember which name they gave to whom. You’ll hear a pause of three or four seconds. A real person just says their name.
"What's your name?" A real person just says it. Someone running several identities needs a moment.
Stage 3: The interviews
Who owns it: the recruiter and the hiring team.
What to look for:
- “My camera isn’t working.” Once happens. The second time is a signal.
- A deepfake check. Ask the candidate to pass a hand in front of their face, turn their head or pick up an object. Overlays can break at the edges. Deepfakes keep getting better, so treat a pass as one data point.
- Answers that sound read. Seconds of silence before every answer, often after repeating the question out loud. Eyes moving across a second screen. A whisper, or someone off camera.
- They can’t defend their own numbers. “Where does the 37% come from?” is the strongest question you have. A real engineer has a messy history: “The migration took nine months. We rolled it back twice.” Someone with a borrowed career has textbook answers and no scars.
- A perfect first answer and an empty follow-up. Ask what went wrong, and what they would do differently.
- No local knowledge. Ask about something local that isn’t in the CV. The FBI recommends soft questions about specific details of a candidate’s location and education for the same reason.
- A different person in the next round. A different voice, face or level of knowledge than at screening. Ask every interviewer for two lines on how the person looked, sounded and answered, and compare them.
On AI: most good candidates use it to write a CV, and that’s fine. Ask openly how they used it, and agree that the live technical part happens without it. Most people respect that. Someone who can’t do the work without a second screen will struggle.
Stage 4: The offer and verification
Who owns it: talent acquisition, working with your ID verification and background check vendors.
This is where most companies run their heaviest checks. They matter, and they work best as the last layer of several.
- ID verification, done properly. Use a verification tool for finalists, written into your hiring policy. See the person live with the document and match them to the person from the interviews. No ID photos in chat or email.
- References you check yourself. Find the former employer on your own and call its main line. One referee name across several candidates is a pattern.
- A background check. For finalists, and earlier for flagged applications. It confirms a history is real. It can’t tell you whose history it is, and stolen identities are often tested against it first. More in Won’t fake candidates get caught at contract signing?
- Audit your vendors. If you hire through staffing firms or outsourcing partners, ask what they check and how. The FBI recommends auditing their hiring practices routinely.
Stage 5: Day one and the first months
Who owns it: IT and the hiring manager.
Some fake candidates get through everything above. The goal now is to find out in week two, long before month eight.
- The address changes right after the offer. The laptop should go to “a cousin”. The FBI tells HR to watch for changes to address or payment details during onboarding. Ship the laptop to the address on the documents.
- Minimal access to start. Expand it over time. Verify identity once more before accounts go live.
- Logins from many places. The FBI warns about many logins into one account in a short time, from IP addresses in different countries. Microsoft reports a spike in impossible travel alerts for these new hires in their first months.
- Remote access tools, VPNs or KVM hardware on a company laptop that IT didn’t set up.
- Working hours that don’t fit the time zone, urgency around payroll setup, or code and files moving to personal accounts.
- The same person, on day one and months later. Compare face, voice and skills with the person you interviewed: on the first team call, then at three and six months. Swaps happen after onboarding too.
If you find one, plan the exit with security and legal before you act. The FBI reports death threats against former managers and their families after termination, and in-person threats against companies that removed these workers’ accounts.
Stage 6: Make it a process
The checks above work only when somebody owns them. Four things turn a list into a process:
- One owner. One person owns candidate fraud in hiring and decides when to escalate. Usually someone in talent acquisition, with a direct line to security.
- A written path. Signal, then a question on the next call, then ID verification for that one candidate. Nobody is rejected on a single flag.
- Friction that grows with commitment. Light and invisible at the application. Heavier checks for the few who reach the offer or earn a closer look.
- A record. Note what was found, what you asked and why you decided. It protects the candidate and you.
Two more habits help. Apply the same steps to everyone: it’s fairer, and fake candidates learn fast which profiles get extra checks. And keep a friendly line ready for the awkward moment: “We verify every candidate this way.” Most people with one odd signal are real, and this keeps the conversation easy.
Bring security in early. Tell them who joined this month and why the first weeks matter, and agree who does what at each stage.
Free tools for the manual checks
Everything here is free and takes a few minutes per application:
- Who made the PDF. Open the document properties in a PDF reader: author, creating app and dates.
- Text you can’t see. Select all in the PDF, or paste it into a plain text editor. White and tiny text shows up.
- Where the photo appears. Run the profile photo through a reverse image search such as Google Lens or TinEye. The same face under other names is a pattern.
- How old the profile is. On LinkedIn, open the profile menu and choose “About this profile” to see when it was created and what was verified.
- Did the employer exist then. Look the company up in its country’s official register, then check its website on the Wayback Machine for the years on the CV.
- Did the code exist then. On GitHub, check when the account was created and whether the commit history matches the years and the stack on the CV.
Want all of this on one printable page, with an owner for every stage? Get the free candidate fraud defense checklist.
What you can start tomorrow, with no budget
- Name one owner. Then write the path from a signal to a decision on one page.
- Share a signal list with the team. Yellow signals: ask more. Red signals: stop and verify. Use the lists above as a start.
- Train hiring managers along with recruiters. They run most of the interviews and make the final call.
- Add a checkpoint at each stage. One or two checks per stage is enough to start.
- Keep a record. What you saw, what you asked, what you decided.
If you want the reasoning behind this plan, read Why fake candidates are rising now.
What we learned running checks on every application
We built TalentShield for our own agency first, and ran it on every application Team Up received before we offered it to anyone else. Four lessons from that month:
- Checks only work if they run on everything. A check a recruiter has to remember is a check that gets skipped on a busy day.
- One signal is a question. A cluster is a pattern. Honest people have messy CVs, gaps and odd file names. Signals from several areas at once tell you much more.
- Real applicants game the ATS too. Some of what we found came from real people with a tool and an incentive, such as hidden text aimed at screening software.
- New patterns keep coming. Whatever you check today, plan to review the list every few months.
Check a little at every stage, starting with the application. Give every check an owner.
Where TalentShield fits
No tool catches everything, and nobody should promise that. Even expensive background checks let operators through.
TalentShield is the first layer. It sits at the top of the funnel, where you have the most applications and manual checking costs the most. It checks every application inside your ATS against 70+ candidate fraud detection signals before anyone books a call, and shows the recruiter what it found and why. Candidates don’t do anything extra. Everything after that, the calls, the interviews, the offer and the first day, still needs people who stay alert.
The bottom line
The manager I spoke with did nothing unusual. That’s the scary part.
Fake candidates will pass the stage you check hardest. So check a little at every stage, give every check an owner, and start with the application.
Book a demo and see what turns up in your own pipeline before the next offer goes out.
Questions people ask
How do you detect candidate fraud in remote hiring?
Check a little at every stage and give each stage an owner. Run automated checks on every application, make a short call before the first interview, run live interviews and compare notes on whether the same person shows up every round, check identity and references at the offer, and watch access and devices in the first months. One signal is a reason to ask. Several signals from different areas are a reason to verify before you go further.
What are the best practices for detecting candidate fraud in remote hiring?
Check every application before anyone reads it. Name one owner for candidate fraud in hiring. Write down the path from a signal to a decision, so nobody is rejected on a single flag. Keep checks light and invisible for everyone at the start, and save heavier checks for finalists and flagged applications. Record what you found, what you asked and why you decided. Apply the same steps to every candidate.
Who should own candidate fraud in hiring?
One person, usually in talent acquisition, with a direct line to security. Under that owner, automated checks and recruiters cover the application and first contact, the hiring team covers interviews, TA covers identity and references at the offer, and IT and the manager cover the first months. The owner decides when to escalate.
What are the most common signs of a fake candidate?
At the application: a CV file whose Author field shows a different name, hidden text aimed at AI screening, pseudo-precise metrics, a VoIP number instead of a mobile, an email that doesn't match the name and a LinkedIn profile created this year next to ten years of experience. Later: a camera that doesn't work twice, answers that sound read, no local knowledge and a different person between interview rounds.
Can you spot a deepfake in a video interview?
Sometimes. Face-swap tools still struggle when a hand crosses the face, when the head turns to a profile view and when the light changes suddenly. Ask the candidate to pass a hand in front of their face, turn their head or pick up an object. Deepfakes keep getting better, so treat a pass as one data point.
Aren't background checks and ID verification enough?
They confirm that a history and a document are real. With a stolen identity, they are. In October 2026 the FBI warned that North Korean IT workers buy accounts with identity verification providers to confirm that stolen identities will pass background checks before they apply. Keep these checks for finalists, and add lighter checks earlier, starting with the application.
What can a TA team start without budget?
Name one owner. Share a one-page list of yellow signals (ask more) and red signals (stop and verify). Train hiring managers along with recruiters. Add one or two checks at each stage, apply the same steps to every candidate and prepare a friendly line for the awkward moment: "We verify every candidate this way."
Sources
- FBI IC3, October 2026: North Korean IT workers expanding use of facilitators, broadening recruitment (PSA261013)
- FBI IC3, January 2025: North Korean IT workers conducting data extortion, with recommendations for employers (I-012325-PSA)
- Nisos, June 2026: inside a DPRK employment fraud operation (166,893 applications, 76 offers)
- Checkr, October 2026: The Hiring Hoax 2026
- HR Dive, August 2025: Gartner, 1 in 4 candidate profiles could be fake by 2028
- Microsoft Threat Intelligence, April 2026: Detection strategies across cloud and identities against infiltrating IT workers
- Microsoft Threat Intelligence, June 2025: Jasper Sleet, North Korean remote IT workers' evolving tactics
- Huntress, August 2026: Insights into suspected DPRK workers, red flags to look out for
- James Reason, BMJ 2000: Human error, models and management (the Swiss cheese model)
- Team Up / TalentShield: first month of TalentShield on live roles, and hiring experience 2024 to 2026 · internal


