When Everything Can Be Generated, Evidence Becomes the Scarce Asset
AI is making professional claims cheaper to produce and harder to trust. The next hiring architecture will not be built around better representations of people. It will need evidence underneath them.
Hiring managers are asking some remote candidates to remove virtual backgrounds, pan their cameras around the room and wave a hand in front of their face. Others are adding live work, screen sharing, device monitoring, IP checks or an in-person stage. What looks like an unusual fraud story points to a larger information problem.
The labour market is losing its ability to treat representation as evidence.
Professional markets have operated on representations.
Market observationThe Wall Street Journal reported in August 2026 that employers were introducing physical camera checks, live assignments, real-time screen sharing and renewed in-person interviewing to establish that remote candidates were who they claimed to be and possessed the knowledge they presented.5
This is evidence of practices among the organisations described, not proof that deception is prevalent across the labour market. The broader implication is Sandery's interpretation.
For decades, employers could usually treat a CV as evidence of experience, an application as evidence of interest, a cover letter as evidence of communication, an interview answer as evidence of knowledge and a take-home exercise as evidence of capability.
None of those assumptions has disappeared. AI has weakened the relationship between the visible output and the behaviour the output was thought to reveal.
When the signal can be manufactured cheaply, the signal carries less evidence about what produced it.
Employers already distrust the résumé.
Industry researchResearch published by iCIMS and Lighthouse Research & Advisory in August 2026 surveyed 463 high-volume and frontline employers. Only one-third were confident that a résumé matched a candidate's skills. Half were concerned about exaggerated or fabricated skills.3
Skills-based hiring ranked as the employers' highest recruitment priority for the following 12 to 18 months. At the same time, résumé and profile screening was the area in which respondents most often said AI was creating value.
That combination is revealing. Employers trust the document less, yet continue investing in technology that processes it.
The next question is not simply what replaces the CV. It is what constitutes sufficiently credible evidence about a person.
Employers confident it matched the candidate's skills.
iCIMS + LighthouseEmployers worried about exaggerated or fabricated skills.
iCIMS + LighthouseSkills-based hiring for the next 12–18 months.
iCIMS + LighthouseAt the extreme, the employer must first establish that the candidate exists.
Academic frameworkA 2026 paper in Human Resource Management Review describes the “phantom candidate”: a synthetic persona constructed using fabricated identities, AI-generated credentials and deepfake technology to obtain remote employment.1
The paper is a conceptual framework addressing an emerging security threat. It does not estimate how common phantom candidates are. Its important contribution is epistemic: a selection system cannot evaluate whether a person will perform if the human referent behind the evidence has not been established.
For most of recruitment history, employers could begin with the question: is this candidate good enough?
AI increasingly requires a prior sequence.
Even demonstrated work now requires two different forms of confidence.
Software hiring offers an advanced version of the problem. Coding assessments were designed to move closer to demonstrated capability. Generative AI then made it possible to produce the output without exposing how the candidate reached it.
HackerRank's July 2026 product release shows two responses occurring simultaneously.4 Its assessment environment evaluates how candidates use AI across IDE activity—how they explore, plan, apply guidance and develop a solution. It also includes integrity controls covering screens, devices, additional faces, suspicious applications and, in limited availability, gaze patterns.
These address different questions.
the observed behaviour?
relate to the actual work?
A system can verify the wrong activity perfectly. It can also observe highly relevant work without knowing who performed it. Trust requires both integrity and validity.
SIOP guidance on algorithmic selection makes the same distinction formally: reliable processing is not enough. The inferences produced must be supported by substantive theory and evidence connecting them to future job performance.2
The more useful response is not to pretend AI does not exist.
Market observationThe Financial Times reported in June 2026 that employers were introducing assessments in which candidates used AI to solve work-like problems.6 McKinsey, for example, added AI to problem-solving interviews for junior consultants. Candidates are assessed on how they research, analyse data, test ideas and exercise judgment over the output—not simply whether they know how to prompt a model.
This is one employer's evolving practice, not a controlled finding that the method predicts performance. But the direction matters.
The assessment moves away from asking whether a candidate can reproduce unaided work that the job itself will no longer require. It moves toward observing how the person performs the augmented work.
That is more sophisticated than replacing credentials with a list of skills. It connects capability to behaviour in context.
The relevant proof is not that somebody can work without AI. It is what remains distinctive when they work with it.
Proof should mean better evidence—not more theatre.
Sandery interpretationThe proof economy should not become an argument for universal proctoring, biometric surveillance or a longer gauntlet of assessments.
Identity checks respond to genuine risks in particular contexts. Applied indiscriminately, they can add friction, intrude on privacy and still reveal little about whether somebody will succeed.
Nor is every professional claim expected to carry documentary verification. Careers contain confidential work, collective achievements, judgment calls and outcomes affected by many people. Evidence has different strengths and practical limits.
The objective is not certainty. It is to make the basis of a claim more visible, preserve provenance where it exists and direct validation toward the uncertainty that actually matters.
because they are available
changes the judgment
A skill label is still a claim.
Skills-based hiring is a rational response to weakening credentials. It remains incomplete if the skill itself is merely asserted.
“Strategic thinker.” Almost meaningless.
“Eight years in corporate strategy.” A proxy.
“Led European market-entry strategy.” A claim.
“Evaluated six countries, developed the investment cases, recommended entry into three and presented the decision to the executive committee.” More specific evidence.
The next layer is the context underneath: what the person owned, the information available, the constraints faced, the alternatives considered, the decision made, the contribution attributable to them and the outcome that followed.
Finally, where the decision warrants it, an employer can present an analogous problem and observe how the person reasons.
- Skill label
- Career proxy
- Specific experience
- Contextual evidence
- Source provenance
- Expert questions
- Observed work
- Structured judgment
The professional claim becomes an interface.
Sandery interpretationProfessional credentials may increasingly divide into two categories: claims, and claims supported by accessible evidence.
“Significant M&A experience” remains useful as a concise interface. The intelligence sits beneath it: the transactions identified, how many completed, the person's role, buy-side or sell-side exposure, relevant jurisdictions, decision authority, contribution across origination, diligence and negotiation, and the provenance and uncertainty attached to those facts.
This does not require publishing confidential information or reducing a person to a score. It requires separating the representation from the system of record.
The same claim can then be expressed differently for an investor, employer, executive-search firm or promotion committee without changing the underlying facts.
The sentence remains the interface. The evidence lives underneath it.
AI weakens the old signals and makes a better architecture possible.
Historically, deeper evidence was administratively expensive. A recruiter could not read a hundred-page career record for 500 people. A partner could not spend three hours assessing everybody who might plausibly succeed.
AI changes the economics of interpretation. It can retrieve the small number of experiences relevant to a particular hiring question, identify where the evidence is direct or transferable, expose gaps and direct a qualified person toward the uncertainty requiring judgment.
That does not make the machine the decision-maker. It makes human attention more selective.
The architecture can move from broad claims and repeated process toward progressively stronger evidence and targeted validation.
- CV
- Recruiter screen
- Repeated interviews
- Offer
- Identity + provenance
- Relevant career evidence
- Expert validation
- Observed work
Build the professional record before the claim is required.
The Sandery viewSandery's current starting point is the individual.
The Career Bank is intended to preserve a richer, longitudinal record of experiences, responsibilities, decisions, achievements, contexts and outcomes. The CV becomes one representation generated from that evidence—not the location where the intelligence lives.
This is not a claim that Sandery can prove capability perfectly or predict performance with certainty. A record can still be incomplete. Evidence can be ambiguous. Capability remains contextual. Human judgment remains necessary.
The longer-term proposition is that better intelligence beneath the professional claim can support better representation, matching, assessment and intermediation. Where the evidence is strong, it should travel. Where it is incomplete, the uncertainty should remain visible.
In a market capable of manufacturing the perfect claim, the valuable institution may be the one whose evidence means something.
The proof economy
When representations become cheap, evidence becomes valuable.
AI does not make every CV false, every interview unreliable or every credential obsolete. It reduces the cost of producing outputs that resemble the traditional evidence of interest, communication and capability.
Rational employers will discount those outputs unless they can understand what sits beneath them.
The scarce asset shifts from the ability to make a professional claim toward the identity, provenance, context and observed behaviour that allow another person to place confidence in it.
September thesis
Stop matching claims.
Build decisions around evidence.
The future advantage is unlikely to belong to whoever can generate the most convincing representation. It will belong to whoever can establish the most useful, trusted and decision-relevant understanding beneath it.
Start beneath the CV
Build the professional record first.
References
Evidence used in this note.
“The proof economy,” the proposed hierarchy from claims to evidence and the broader architecture are Sandery interpretation. The sources establish specific changes in employer confidence, identity verification and assessment practice; they do not demonstrate that every professional signal is unreliable or that the proposed sequence operates uniformly across roles and markets.
Research & institutional evidence
- The phantom candidate: A framework for combating AI-enabled identity deception in remote hiringPankaj C. PatelHuman Resource Management Review · Framework addressing synthetic identities and ontological validity in selection, 2026 ↗
- Algorithmic JusticeSociety for Industrial and Organizational PsychologyGuidance distinguishing reliable processing from evidence supporting job-related inferences, 2020 ↗
Industry research
- The New Reality of Frontline Hiring in 2026iCIMS and Lighthouse Research & AdvisorySurvey of 463 high-volume and frontline employers, published August 2026 ↗
- July 2026 Release NotesHackerRankProduct documentation covering AI-fluency evaluation, observed IDE activity and assessment-integrity controls ↗
Market reporting
- Employers Are Making Job Candidates Jump Through Hoops to Prove They’re RealThe Wall Street JournalMarket reporting on identity checks, live work and renewed in-person verification, August 2026 ↗
- Job interviews are becoming AI testsFinancial TimesMarket reporting on employers, including McKinsey, assessing AI use through work-like problems, June 2026 ↗