On January 27, GoodTime published its fifth annual Hiring Insights Report with an uncomfortable pairing for talent leaders: almost everyone was moving toward AI agents, and fraudulent or AI-assisted candidates had become the top anticipated hiring threat for 2026.

In the report, 99.8% of talent acquisition teams were using, piloting, or planning to use AI agents. Ninety percent of companies missed their hiring goals, and one in three missed by a wide margin. Recruiters were still spending 38% of their time on scheduling, and 60% of organizations saw time-to-hire get worse in 2025.

The conflict is practical, not philosophical. Companies want AI because the funnel is too large, too slow, and too expensive to run manually. They also need to prove that the person moving through that same funnel is real, qualified, and fairly evaluated.

Old recruiting budgets separated those problems. Scheduling automation sat in one line. Assessment vendors sat in another. Background checks came later. Fraud review was an exception. Candidate experience was measured after the process. Legal review entered when something went wrong.

That separation is breaking.

A candidate can use AI to tailor a resume, generate interview answers, complete a take-home assignment, polish a portfolio, or appear in a video meeting through an avatar. A recruiter can use AI to screen, summarize, schedule, assess, and explain. An ATS can open live hiring data to AI assistants through MCP. A high-volume hiring workflow can compress search, application, screening, scheduling, offer, and onboarding into a few text conversations.

By June 2026, proof of person no longer sits at the edge of hiring as a background-check problem. It now sits inside every stage where the company turns a candidate claim into a hiring signal and decides how much evidence to buy.

January 27 Put Fraud Beside Scheduling

GoodTime’s report matters because it puts candidate authenticity next to operational capacity, rather than treating fraud as a separate security problem.

In the same data file, hiring teams missed goals, scheduling still taxed recruiter time, nearly all teams were turning toward AI agents, and fake or AI-assisted candidates ranked as the top expected threat. The report’s blog excerpt adds a smaller but useful detail: 23% of talent acquisition leaders cited fake or AI-assisted candidates as a current challenge, while the issue ranked as the leading anticipated threat for the year ahead.

The sequence turns authenticity into a budget problem.

If hiring teams were flush with recruiter capacity, they might absorb more verification manually. They could add extra phone screens, ask deeper follow-up questions, check references earlier, or route suspicious files to a specialist. But GoodTime’s data points in the opposite direction. Teams are short of time, hiring goals are still hard to reach, and the fastest teams are not simply adding headcount.

They are reorganizing around automation.

Automation can make the proof problem worse before it makes it better. AI scheduling reduces coordination work, but it also moves candidates faster into interviews. AI screening can process volume, but it may promote cleanly written claims before a human knows whether the claims are true. AI interview summaries help hiring managers read faster, but they can flatten uncertainty. AI candidate communication can make the process feel responsive, while the file underneath still lacks proof.

Recruiting has always lived with embellishment. Candidates round up titles, overstate project ownership, polish accomplishments, and choose references who will support the story. The difference in 2026 is scale and surface quality. A weak candidate can produce a fluent file. A strong candidate can look indistinguishable from a polished weak one. A fraud operation can send more applications than a recruiter can investigate.

Assessment integrity has widened as a result.

The term once described proctoring, coding-test leakage, proxy interviews, or cheating detection. It now covers a wider cost stack: identity verification, claim extraction, work sample design, live follow-up, fraud triage, interview recording, transcript review, candidate transparency, human review, adverse-action evidence, and post-hire remediation if a bad file becomes a bad hire.

None of those steps are free.

A buyer has to choose where the proof happens. Early proof is cheaper when it filters obvious risk before manager time is spent. Early proof is also more legally sensitive if the signal shapes access to an employment opportunity. Late proof may be safer in some jurisdictions because background checks and identity steps often sit after conditional offers. Late proof is costly when fraud has already consumed interviews, manager calendars, candidate communication, and onboarding work.

Budget has to follow the workflow, not the old vendor category.

Resume Claims Require a Proof File

A resume screen used to be the first rough sorting mechanism. It still matters, but it now functions more like a claim file than a decision file.

The ICIMS and Aptitude Research AI adoption report released on April 30 gives the demand-side context. Seventy-four percent of companies said candidates are using AI in the job search. Sixty-nine percent of companies were already using AI in talent acquisition in some capacity, though only 18% were using it broadly across hiring processes. Screening was the leading use case at 58%, followed by candidate communication at 54%, assessments at 50%, and sourcing at 46%.

Recruiters remain central in that report. When AI and humans conflict, recruiter judgment overrides AI recommendations in 58% of organizations. Eighty percent of organizations said recruiters were spending more time engaging and nurturing candidates, 73% were strengthening hiring-manager partnerships, and 64% were spending more time on strategic talent planning and advising.

That sounds like the right human role. It also creates a harder evidence duty.

If a recruiter is the person who overrides the AI recommendation, the recruiter needs a file that explains what they saw. A resume score is not enough. A keyword match is not enough. A fluent cover letter is not enough. A portfolio link may not be enough if the work cannot be tied to the person. A generated summary may be useful, but only if the source records remain available.

The proof file starts with a simple split:

Candidate artifactWhat it can showWhat it cannot show aloneProof step
ResumeClaimed work history, titles, dates, keywords, credentialsReal ownership, judgment, collaboration, authenticityExtract claims and verify the claims that matter for the role
Cover letter or screening answerMotivation and role framingWhether the candidate wrote it or understands itAsk targeted follow-up tied to specific claims
Portfolio or projectWork product and technical vocabularyIdentity, contribution level, team contextRequire explanation of choices, constraints, and tradeoffs
AI interview transcriptResponse to a consistent promptWhether the prompt measured the right skill or captured nuancePreserve rubric version, transcript, recording policy, and reviewer notes
Work sampleObservable performance on a taskLong-term performance or full identity proofPair with live discussion and identity checkpoint
Fraud signalRisk triggerJob ability or lawful rejection basisRoute to human review and document the evidence basis

At first glance, the table looks operational. In practice, it is a budget map.

Every cell costs money. Claim extraction may come from the ATS, an AI assistant, or a recruiter. Verification may come from a background-check vendor, credential service, reference process, or work-sample review. Live follow-up consumes recruiter or hiring-manager time. Fraud triage may require a specialist queue. Candidate transparency requires communication design and recordkeeping. If the process reaches an adverse outcome, the employer may need explanation and reconsideration workflows.

The wrong budget posture is to buy one more AI screener and call the proof problem solved.

A screener can rank claims. It does not make the claims true. A fraud detector can flag suspicious behavior. It does not prove job competence. A structured interview can produce comparable answers. It does not prove the person did the take-home assignment. A background check can verify identity and some records. It often happens after earlier stages have already consumed time.

Hiring teams need a proof sequence that fits role risk. A warehouse associate, a nurse, a cleared engineer, a remote software contractor, a finance analyst, and a customer support representative should not have identical proof plans. The job risk, access level, hiring speed, candidate supply, local law, and fraud pattern should decide where the organization spends.

For that reason, assessment integrity is moving from a vendor feature into hiring operations.

GCheck Shows Where Preparation Crosses Into Impersonation

GCheck’s 2026 Trust in Hiring Report gives the candidate-side tension. The report is based on a national survey of 1,500 U.S. adults employed full-time who had actively applied for at least one job in the past 18 months. It found that candidates are using AI across a spectrum that runs from ordinary preparation to active misrepresentation.

The numbers are blunt. In the report, 61% said they used AI to practice interview answers until they sounded more impressive than authentic. Fifty-four percent used AI to write a cover letter. Fifty percent tailored a resume without meeting the requirements. Forty-eight percent used AI to complete take-home assignments. Forty-three percent generated overstated resume bullets. Forty-two percent wrote untrue application answers. Twenty-seven percent used AI during a live interview for real-time answer generation. Twenty-five percent used an AI-generated avatar of themselves in a virtual job meeting.

Those behaviors do not belong in one bucket.

Practicing interview answers with AI is not the same thing as using an avatar in a virtual meeting. Writing a clearer cover letter is not the same thing as having another person assist with a technical assessment. Tailoring a resume to emphasize relevant experience can be legitimate. Tailoring a resume to claim requirements the candidate does not meet is different.

Hiring teams need a taxonomy before they need another detector.

Candidate behaviorLikely categoryEmployer response
Uses AI to practice answersPreparationAccept, but design follow-up that tests understanding
Uses AI to rewrite a truthful resumeAssistanceAccept, then verify critical claims
Uses AI to complete a take-home assignmentAssessment contaminationRequire live explanation, version history, or supervised follow-up
Uses AI during a live interview for real-time answersMisrepresentation riskDisclose rules, monitor where lawful, route flagged cases to review
Uses AI avatar in a virtual meetingIdentity riskRequire proof-of-person checkpoint and human review before decision
Has another person assist with a technical assessmentProxy-performance riskRedesign assessment and document review path

The distinction protects candidates as well as employers.

GCheck’s report also shows why candidates want transparency even when they use AI themselves. Eighty-two percent wanted a clear explanation of what is being checked. Eighty-one percent wanted human review of findings rather than fully automated decision-making. Seventy-seven percent wanted the ability to review or dispute findings. Seventy-four percent wanted transparency about AI use in screening.

For employers, those conditions warn against turning proof of person into hidden surveillance.

If employers respond to candidate AI use by adding opaque detectors, silent fraud scores, unexplained video analysis, or undisclosed automated review, they may rebuild one trust problem while creating another. Candidates who use AI for preparation may feel treated like suspects. Candidates from groups already modifying how they present themselves may see more scrutiny without clearer standards. Recruiters may receive a risk score without enough evidence to know whether it is a serious concern or a false positive.

A stronger operating model is disclosed and staged.

Tell candidates which parts of the process allow AI assistance, which parts require unaided work, which identity checks apply, how work samples will be reviewed, what data will be retained, and how a candidate can dispute a finding. Then build assessments that make real understanding visible. A live explanation of a submitted work sample often reveals more than an AI-detection score. A role-specific scenario that asks the candidate to make tradeoffs under follow-up pressure is harder to outsource than a generic take-home prompt.

Disclosure and staging do not eliminate fraud. They lower the chance that the employer pays for the wrong proof.

Checkr Finds the Fraud Too Late

Checkr’s 2026 State of Screening Compliance Report shows where the current process leaks money.

The report says 58% of surveyed HR teams detected hiring fraud in the past year. It lists common types: 34% encountered resume or credential fabrication, 29% document fraud, 20% interview fraud such as proxy interviews or AI-assisted impersonation, and 15% identity fraud involving stolen identity or fabricated personal details.

Timing is the expensive part. Checkr says 38% of teams most often identify fraud during background-check adjudication, 35% during resume or application screening, 18% during interviews, 13% during dedicated fraud-detection or identity checks, 13% during I-9 or identity verification at onboarding, and 13% post-hire through payroll, tax, or onboarding documents.

That means much of the discovery happens after the candidate has already consumed process.

Checkr also makes a practical point about Ban the Box rules. In some jurisdictions, background checks come after interviews or conditional offers. That can be fair policy. It also means fraud controls that rely only on background-check timing may arrive after the team has spent recruiter time, manager time, scheduling work, assessment review, offer work, and candidate communication.

The timing pushes proof-of-person budgeting toward earlier non-background signals.

Earlier does not have to mean invasive. It can mean checking whether the candidate understands their own work. It can mean verifying a required license before a late-stage interview if the role lawfully requires it. It can mean asking for a short live explanation of a take-home assignment. It can mean requiring identity confirmation at the point where a remote interview becomes a final-round decision, subject to local law and job relevance. It can mean using anomaly signals as review triggers, rather than automatic rejection reasons.

Delay makes the budget split visible:

Discovery pointCost already incurredBetter spend upstream
Application screeningLow recruiter time, low manager timeClaim extraction and risk-based follow-up
First interviewScheduling time, recruiter screen, candidate communicationClear AI-use rules and structured follow-up
Technical or work sample stageAssessment design, reviewer time, manager attentionLive review, version evidence, role-specific prompt design
Background check adjudicationInterviews, offer work, legal/compliance handlingEarlier credential gating where lawful and job relevant
I-9 or onboardingOffer, onboarding, provisioning, payroll setupIdentity checkpoint before system access and start logistics
Post-hireSecurity access, payroll, customer data, team dependencyPost-hire monitoring, access controls, and rollback runbook

Worst case is not only hiring the wrong person. It is discovering the wrong person after provisioning accounts, assigning customer work, sending payroll data, or exposing internal systems. At that point, the cost stack leaves recruiting and moves into security, finance, legal, and operations.

Recruiting teams will not solve that with a single vendor. The proof plan needs a sequence of controls with named owners.

TA operations owns process design. Recruiters own evidence review. Hiring managers own job-relevant validation. HR compliance owns notice, record retention, and adverse-action workflow. Security owns access timing and identity risk. Procurement owns vendor obligations when a proof step fails. Finance owns the combined cost of screening, rework, fraud tools, and bad-hire remediation.

Candidate fraud turns hiring into a shared-risk workflow.

Most hiring budgets are not built for that view. Background check spend sits with one owner. Interview scheduling sits with another. Assessment spend sits somewhere else. AI agent spend may sit in a platform credit pool. Manager time often has no line item. Candidate drop-off appears later as sourcing spend. Fraud investigation time may be invisible.

The proof-of-person budget has to make those costs visible together.

Greenhouse and Workday Move Verification Into Live Workflow

Platform vendors are moving faster than the proof layer.

On May 7, Greenhouse announced Greenhouse MCP, with rollout beginning in June. The company described it as a permission-aware way for approved AI tools to connect directly to Greenhouse. It highlighted audit trails, organization-level controls, rate and safety limits, and a curated set of MCP tools. Greenhouse also said 30% of active job seekers in its survey were already using AI agents to search for openings, submit applications, and schedule interviews.

The product direction is clear. AI will not sit outside the ATS waiting for exports. It will operate against live job, candidate, pipeline, offer, and hiring data.

Workday is pushing from another side. Its Q1 FY27 prepared remarks said the Workday Recruiting Agent supported 14 million hiring processes in the quarter, up 44% year over year. Workday also said it had over 80 million users under contract and about 1.4 trillion transactions annually, a data foundation it frames as a model of work for agents. In January, Workday made Paradox Conversational ATS available through Workday, saying candidates could search, apply, interview, and onboard through short text conversations, with a 72% average application completion rate and an average time-to-hire of three and a half days among Paradox customers.

These product moves are useful. They reduce friction. They also compress the time available to detect weak proof.

A conversational ATS can move a frontline candidate from interest to interview to offer quickly. That is valuable when stores, restaurants, warehouses, and care facilities need people. It also means a bad signal can travel quickly through screening, scheduling, offer, onboarding, and first-shift readiness. A governed MCP server can let an assistant summarize pipeline risk or create compliance-ready narratives. It can also turn a weak input into a confident summary if the source file lacks proof.

Verification now has to live inside the workflow instead of after it.

That does not mean every candidate should face heavy identity screening at the first click. It means each workflow should know its proof gates:

WorkflowSpeed promiseProof riskRequired gate
High-volume frontline hiringApply and schedule in minutesIdentity, availability, required credential, first-shift readinessEarly eligibility claim check, role-relevant identity checkpoint before offer or start
AI voice interviewStructured first-round conversation at scaleDisclosure, transcript accuracy, avatar or coaching riskCandidate notice, transcript retention, rubric version, human review route
MCP hiring assistantLive pipeline analysis and action supportSource compression, permission drift, write-action riskSource links, audit log, permission boundary, write-action approval
Technical assessmentSkill evidence without full interviewer loadTake-home contamination, proxy help, prompt leakageLive follow-up, version history, job-relevant task redesign
Candidate fraud detectionFaster risk flaggingFalse positives, opaque scoring, discrimination riskHuman review, evidence basis, dispute path, consistent standards

Vendors will compete on how well they embed these gates without ruining speed.

Vendors face a hard product problem. A high-friction proof process can drive away legitimate candidates. A low-friction process can let organized fraud reach interviews. A detector-heavy process can create fairness and transparency concerns. A trust-heavy process can leave hiring managers vulnerable to polished false signals. The winning product will not be the strictest one. It will be the one that spends proof at the right moment.

Procurement changes too. Buyers should stop asking only whether a vendor has fraud detection. They should ask where the vendor’s proof signal enters the workflow, what it costs, what it blocks, what it preserves, which records are exportable, how a human reviewer sees it, and whether the vendor can distinguish preparation from impersonation.

Proof spending should sit beside speed metrics, not underneath them.

Colorado Makes Proof Reviewable

Regulation turns proof from an internal quality issue into a reviewable record.

Colorado’s SB26-189, signed on May 14, defines automated decision-making technology as technology that processes personal data and generates output used to make, guide, or assist a decision about an individual. Consequential decisions include access, eligibility, or compensation related to employment. Starting January 1, 2027, developers of covered ADMT must provide deployers with technical documentation covering intended uses, training-data categories, known limitations, instructions for appropriate use, and human review. Developers and deployers must retain records needed to show compliance for at least three years. Deployers must provide point-of-interaction notice and, after an adverse outcome, a plain-language description of the covered ADMT’s role within 30 days. Consumers can request personal data, correction of factually incorrect data, meaningful human review, and reconsideration.

California already moved on another record line. The California Civil Rights Department said its employment automated-decision rules went into effect on October 1, 2025, and require employers and covered entities to maintain employment records, including automated-decision data, for at least four years. New York City’s Local Law 144 continues to require bias audits, public audit information, and candidate or employee notice before covered automated employment decision tools are used.

These rules do not ban AI in hiring. They raise the cost of unexplained automation.

For assessment integrity, the key word is “reviewable.” A proof signal that cannot be reviewed may be useful for internal triage, but it is weak as a decision basis. A fraud flag with no explanation cannot carry an adverse outcome. A work-sample score without the prompt, rubric, reviewer note, and candidate response cannot support meaningful human review. An AI interview summary without the transcript and disclosure record is an incomplete file. An MCP-generated recommendation without source links and permission logs is a governance gap.

An employer needs a proof packet, not a pile of tools.

That packet should answer:

  1. Which claims did the candidate make?
  2. Which claims mattered for the role?
  3. Which AI tools or automated systems processed the claims?
  4. Which outputs materially influenced a decision?
  5. Which human reviewed the output and had authority to disagree?
  6. Which records support the assessment, fraud review, or identity checkpoint?
  7. Which notice, correction, dispute, or reconsideration route was available?

A packet also helps recruiters. If a candidate appeals, the recruiter should not have to reconstruct the workflow from five dashboards. If a hiring manager asks why a candidate was removed, the recruiter should be able to point to the evidence. If legal asks which automated step mattered, the answer should not be a Slack thread.

At that point, proof-of-person spending can become strategic. A company that invests in clear proof packets can move faster with less downstream fear. A company that buys disconnected screening, interview, fraud, and AI-assistant tools may end up with more signals and less certainty.

A record also protects against overcorrection. If every AI-assisted candidate becomes suspicious, hiring will punish common preparation behavior and narrow the funnel in unfair ways. If every fraud signal is treated as decisive, the employer may automate exclusion through a proxy it does not understand. A reviewable proof packet forces the organization to separate the candidate’s use of AI, the candidate’s identity, the candidate’s skill, and the candidate’s lawful eligibility.

Those are different questions. The budget should not blur them.

Work Samples Become a Budget Decision

The cheapest proof is not always the best proof.

A generic coding test can be cheap to administer and easy to score. It can also leak online, invite AI completion, and fail to measure the work that matters. A one-way video interview can be cheap to schedule. It can also trigger candidate distrust, avatar risk, and accessibility questions. A heavy proctoring system can deter fraud. It can also create friction for legitimate candidates and add bias or privacy concerns. A full in-person final round can confirm identity and interaction quality. It can also slow hiring and disadvantage candidates who cannot travel.

A buyer has to price the whole sequence.

Spend categoryWhat it buysHidden cost if overusedHidden cost if missing
Identity verificationConfidence that the candidate is a real person tied to the filePrivacy friction, candidate drop-off, jurisdictional complexityProxy interviews, stolen identity, post-hire risk
Proctoring and monitoringControlled assessment conditionsCandidate distrust, accommodation burden, false positivesAI coaching, outside help, leaked answers
Work-sample redesignBetter job-relevant signalManager/reviewer time, scoring calibrationGeneric tests that reward surface fluency
Live follow-upProof that the candidate understands their own workInterviewer capacity, scheduling loadTake-home contamination, portfolio ambiguity
Fraud investigation queueHuman review of risk flagsSpecialist staffing, SLA pressureSilent rejection risk, unreviewed fraud patterns
Candidate transparencyRules, consent, dispute path, trustCommunication design, legal reviewWalkouts, complaints, weak evidence
Evidence exportReviewable record for HR, legal, and auditorsVendor implementation cost, storage, data governanceNo defense when a decision is challenged

Those spend lines explain why “assessment integrity” will not be a single market.

Some buyers will spend on identity verification and background screening. Some will spend on interview intelligence and structured scorecards. Some will spend on work-sample platforms. Some will spend on proctoring. Some will spend on fraud detection. Some will rebuild internal assessment design and use vendors only for plumbing. High-volume frontline employers will favor fast, mobile proof gates. Technical employers will favor work samples and live explanation. Regulated or safety-sensitive employers will need more identity, credential, and recordkeeping controls. Remote global employers will need stronger proof before system access.

A mature buyer will not ask, “Which tool catches cheating?” It will ask a more specific set of questions:

  • Which stage is currently producing the most false confidence?
  • Which proof step would reduce manager time wasted on weak files?
  • Which proof step creates legal or candidate-experience risk if moved too early?
  • Which candidate behaviors are allowed, discouraged, or disqualifying?
  • Which human has authority to review a fraud or assessment flag?
  • Which records remain exportable after the vendor contract ends?
  • Which cost should be paid by HR, security, finance, or the hiring business?

The last question matters because proof benefits more than recruiting.

Security benefits when a fake candidate does not receive system access. Finance benefits when payroll and onboarding errors are avoided. Legal benefits when adverse decisions have records. Hiring managers benefit when interviews test real capability. Candidates benefit when the rules are clear and honest work is not buried under AI-polished volume. Procurement benefits when vendor claims can be tied to evidence.

Expense should be shared accordingly.

Vendors have a fair objection. No assessment product can prove perfect job performance. No fraud signal can guarantee intent. No identity check can show whether a person will succeed after a manager changes the role. Employers write vague job descriptions, move criteria late, and sometimes use assessment tools to avoid doing calibration work. A vendor should not be held responsible for every bad hire.

That objection is correct. It should make the contract sharper.

Vendors should be accountable for the evidence they sell. If they sell identity assurance, the signal, scope, and failure mode should be documented. If they sell AI interview scoring, the prompt, rubric, transcript, review process, and candidate communication should be exportable. If they sell work-sample assessment, the task should map to job-relevant skills and reviewer calibration. If they sell fraud detection, flagged cases should include an evidence basis and a human-review workflow. If they sell AI summaries, source links and uncertainty should survive the summary.

Employers own the hiring decision. Vendors own the integrity of their proof objects.

That boundary is where the market will settle.

Ninety Days Later, the Hiring Manager Checks the Badge

After the hire, the proof budget becomes clearest.

Imagine a remote support analyst hired in June. The resume matched the role. The AI screen looked strong. The take-home exercise was polished. The video interview was smooth. The background check cleared. The candidate started on time. Ninety days later, the manager notices that the employee cannot explain the troubleshooting steps in their own submitted assignment. A security review finds unusual access patterns. Payroll asks why onboarding information does not match another record. The recruiter opens the candidate file and sees a collection of clean summaries, but not enough proof to know where the signal failed.

That version is costly because it is no longer only a recruiting miss. It touches access, payroll, customer data, manager trust, team morale, vendor accountability, and maybe a compliance record.

Now imagine the same file with a different proof design. The resume claims were extracted and marked by importance. The take-home exercise required a live explanation and reviewer notes. The AI interview transcript and rubric version were retained. A candidate-facing notice explained AI use and the review route. A fraud signal triggered a human review rather than silent exclusion. The identity checkpoint occurred before sensitive access. The final packet showed what the recruiter, hiring manager, and tool each contributed to the decision.

That file still cannot guarantee a good hire. It can show where the organization made a reasonable decision.

Proof of person is not a slogan. It is a sequence of spending choices that decide how much uncertainty a company is willing to carry from application to start date.

Companies that handle it well will not be the ones that suspect every candidate. They will be the ones that define acceptable AI use, test job-relevant work, verify identity at the right moments, preserve reviewable evidence, and keep humans accountable for the judgments that matter.

The hiring funnel is filling with AI on both sides. A better budget meeting starts when someone asks a simpler question: before we spend another interview hour, what do we actually know about the person in the file?


This article provides a deep analysis of candidate fraud, assessment integrity, and proof-of-person budgeting in AI-assisted hiring. Published June 4, 2026.