In this article
Hiring teams need to verify relevant claims and identity through a consistent, proportionate process. They also need to avoid turning every nervous applicant into a suspected deepfake. These goals belong together. A fair process is easier to explain and investigate than a collection of improvised tests invented during an awkward video call.
Greenhouse's September 2026 candidate fraud guide describes impersonation, synthetic identities and AI-assisted deception. It is vendor guidance, not proof that a particular applicant is fraudulent. Its 2026 US hiring survey also captures the wider trust problem. Candidates and employers both face uncertainty about what is authentic and what the other side will do with their information.
Separate three different questions
Is the person who they claim to be? Can they do the work? Did they follow the assessment rules? These questions require different evidence. A legitimate person can lack the relevant skill. A skilled person can misunderstand unclear AI rules. An identity check cannot tell you whether someone makes good product decisions.
State the assessment policy in advance. If AI help is permitted for a take-home exercise but not a particular live task, say so. Translation tools, accessibility support and ordinary preparation should not be quietly reclassified as deception after the interview. Clarify the boundary before evaluating behaviour against it.
Treat uncertainty as a reason to investigate fairly, not an excuse to invent certainty. A failed video connection, an unfamiliar accent or a delayed answer may have entirely ordinary explanations. None should function as a shortcut for deciding that someone is dishonest.
Test understanding through the work
Use a short role-related scenario and ask the candidate to explain a decision. Introduce a new constraint, invite questions and discuss the trade-off. Someone who understands the work should be able to engage with the reasoning, even if they need time to think.
In a hypothetical product interview, ask how a proposed rollout changes when customer data cannot leave an approved environment. This reveals judgment about constraints. It does not require the interviewer to guess whether a pause means a chatbot is answering. CIPD's selection guidance is a stronger basis for assessment design than trying to read dishonesty from facial expressions.
Do not use automated emotion, personality or honesty judgments from a face or voice as evidence of competence or fraud. Avoid uncertain AI-text detection as a verdict on authorship. Evaluate the substance, seek clarification and use appropriate verification for material claims.
Verify identity at the appropriate stage
Agree with HR and security what verification is necessary, when it occurs and who handles it. Use a secure, approved process and collect only what is required. The appropriate method depends on the role, jurisdiction and risk; not every early screening call justifies collecting identity documents.
Explain the reason, the data involved, retention arrangements and a contact for questions. Provide a reasonable alternative where a technical method fails or an adjustment is needed. A genuine applicant should not have to email sensitive documents to an individual interviewer because the meeting became uncomfortable.
Identity verification and right-to-work checks can involve distinct legal requirements. Let the qualified owners define them. An operational hiring guide cannot replace country-specific advice or the company's formal process.
Investigate discrepancies without a public accusation
Record the specific inconsistency: two conflicting employment dates, a claimed qualification that cannot be confirmed or a material difference in the account of a project. Give the candidate a chance to explain. Memories, job titles and administrative records can be messy without being fraudulent.
Use references through approved channels and with the necessary permissions. Ask about relevant work rather than fishing for unrelated personal information. If something suggests a security threat, involve the team equipped to handle it; recruiters should not be expected to improvise incident response between interviews.
Document evidence and decisions separately from suspicion. “The candidate could not explain the assessment's central calculation” is assessable evidence. “They seemed too polished” is an impression that needs much more care.
Protect candidates from fake employers too
Trust works in both directions. Send interview information from verifiable company channels, identify interviewers and explain the stages. Make the real vacancy easy to find on the official careers page. Candidates should not have to guess whether a request for information comes from the company or somebody impersonating it.
Avoid asking applicants to install unfamiliar software without a clear reason and review. Explain which systems will be used. Maintain a route for reporting suspicious contact. The company's hiring process should model the verification habits it expects from candidates.
Cisco's privacy benchmark is useful further reading on responsible data handling. The practical implication here is to make hiring data part of governance, rather than treating the recruitment folder as an exception because everybody is in a hurry.
Check whether the defence is blocking good people
Review false alarms, delays and withdrawals as well as confirmed incidents. Ask whether an adjustment route works and whether the same standards are applied across applicants. A control that creates friction for everybody while providing little evidence may be expensive reassurance.
Keep the process short enough to operate consistently. Named ownership, clear AI rules, structured assessment and proportionate verification create a better foundation than a recruiter maintaining a private catalogue of suspicious blinking patterns.
For the assessment itself, read how to hire from work samples. If software screens candidates before a person sees them, continue with human oversight of AI screening. I help teams turn messy cross-functional workflows into decisions people can explain. Bring the point where your hiring process loses trust, and see my background for the practical work behind these articles.
Questions people ask
How can employers detect deepfake interviews fairly?
Use a consistent process combining job-related assessment, clarification of discrepancies and proportionate identity verification through approved channels. Do not treat a lagging video, accent, disability or nervous behaviour as proof of fraud. Where concerns remain, involve HR and security and give the candidate a fair opportunity to explain.
Does using AI in an interview mean a candidate committed fraud?
It depends on the agreed rules and what was represented. Allowed assistance, accessibility support and deliberate impersonation are different. State the policy before assessment and test understanding of the work. Avoid relying on an uncertain detector or an interviewer’s impression as a conclusive finding.
Should every applicant submit identity documents?
Not automatically at the first contact. Determine the necessary stage and method with the qualified HR, security and legal owners. Collect only the required information, use secure channels, explain retention and provide an appropriate route when verification fails or an adjustment is needed.
Something here sounds like your workplace?
AI integration, workflow automation, team training or a product leadership conversation.