Four surveys published in the past four months, from an accountancy body, a skills charity, an HR platform, and a hiring vendor, all point at the same fracture line in HR tech. Employers are deploying AI screening and interviewing tools faster than they are building the trust, or the skill, to use them well. The result is not a single scandal but a slow accumulation of evidence that the industry’s AI hiring stack is running ahead of the humans who are supposed to govern it.
The Adoption Curve Outran the Trust Curve
Start with the numbers. In the Association of Chartered Certified Accountants’ Global Talent Trends 2026 survey of more than 11,000 professionals across 160 countries, 48% of respondents said they have reservations about the use of AI algorithms in hiring processes. That skepticism is not concentrated among junior staff who feel threatened by automation. It is heaviest at the top: 54% of board-level leaders told ACCA they have doubts about relying on AI to select talent, a higher share than among the workforce as a whole.
That is a striking split. The people with the authority to slow down or redesign an AI hiring rollout are, on average, more wary of it than the people beneath them who actually run it day to day. If board-level confidence is the binding constraint on how aggressively HR can automate screening, then a near-majority of boards quietly doubting the system is not a rounding error. It is a governance gap sitting one layer above the recruiting function.
ACCA’s own read of the data cautions against treating that 48% as a single monolithic number: “Attitudes vary considerably across countries, sectors and by generation,” the report notes, pointing out that professionals in accountancy practice and younger, Gen Z respondents tend to view AI recruitment more favorably than the workforce overall. That regional and generational spread matters for HR leaders operating across multiple markets: a policy calibrated to the most skeptical board in the portfolio may be unnecessarily conservative for a workforce that is, on average, considerably more comfortable with the technology than its directors are.
What the Tools Are Actually Screening For
The board-level unease tracks with what is happening inside the tools themselves. City & Guilds’ September 2026 report, “Counting the cost: The UK’s Employability Skills Crisis,” found that 69% of employers believe AI applicant screening prioritizes technical qualifications over the human skills, communication, problem solving, time management, that they say they actually need. More than a third of employers said the resulting skills shortage is already showing up as operational errors and delays, and a similar share pointed to higher turnover and lower morale.
Andy Moss, interim chief executive officer of City & Guilds, put the mechanism plainly: “Employers are telling us loud and clear that employability skills, the ability to communicate, solve problems and manage time well, matter just as much as technical know-how, if not more. Yet at the exact moment these skills are most in demand, we’re seeing a growing human skills shortage, made worse by AI-powered recruitment tools that can filter out exactly the candidates businesses need most.”
That is not a complaint about AI screening being unfair in the abstract. It is a specific claim that the optimization target is wrong: the tools are tuned to match keywords and credentials against a job description, and the employability signals that determine whether someone is actually good at the job rarely survive that filter. HR technology built to speed up screening can end up narrowing the funnel in a direction the employer did not intend.
City & Guilds’ data also points to where the cost eventually lands. More than 1 in 3 employers in the report linked the employability skills shortage directly to productivity problems, and separately to higher turnover and lower morale. On the candidate side, 51% of 18 to 24 year olds told researchers they worry AI will reduce job opportunities for people their age, a fear the report ties to the more than 1 million young people in the UK who are currently not in employment, education, or training. The report’s other finding, that apprentices report the highest confidence in face-to-face communication skills among their peer group, is the closest thing in this research to a fix already in evidence: routes that combine practical experience with structured human interaction are producing exactly the skills employers say their AI-filtered pipelines are missing.
The Taxonomy Problem Underneath the Screening Problem
Part of why the filters misfire traces back to a more basic confusion: employers and vendors often cannot agree on what an AI-related job even is. Research published by Andela on September 10 analyzed roughly 1,832 postings titled “AI Engineer” or “ML Engineer” and found that 53% of them requested skills drawn from at least two genuinely different established roles. The firm’s methodology also surfaced 23 recurring skill bundles across Fortune 500 hiring that do not map to any standardized job title, including 6,758 postings that carry the skill bundle of an “LLM application engineer” without ever naming the role.
Cory Hymel, Andela’s head of research, described what that mismatch costs employers: “Any gap between the role a company thinks it’s hiring for and the role the work requires becomes a structural liability. Our methodology of reading skills inside job postings enables us to see the gaps so customers and talent can better fill them.” Andela chief executive Carrol Chang framed the broader stakes the same way: “AI is changing tech workflows faster than ever, and we need new ways to get ahead of those changes. This groundbreaking research, based on our proprietary skills taxonomy updated for the AI world, provides a headlight, not a rear view mirror.”
Put together with the City & Guilds finding, the picture is a hiring stack confused at two levels simultaneously: the job descriptions going in do not cleanly describe the roles being hired for, and the screening tools reading those job descriptions are tuned to filter on the wrong signal once they arrive. Neither problem is a bias story in the sense regulators usually mean. Both are closer to a specification failure, and specification failures are exactly the kind of defect that compounds silently until an audit, a lawsuit, or a bad quarter of attrition surfaces it.
Andela’s research also identified a more constructive signal buried in the same data: eight genuinely new roles and 14 hybrids of old and new roles that its skills taxonomy could name even though standard job-title libraries could not. For HR technology vendors, that is a direct product problem: an applicant tracking or HCM system whose skills taxonomy has not been refreshed for the AI-era job market is, by Andela’s own count, misreading well over half the AI-adjacent postings that pass through it. The fix is not a smarter algorithm sitting on top of a stale taxonomy. It is refreshing the taxonomy itself.
What Candidates See From the Other Side of the Screen
If the employer side of the trust gap is about whether the tools work as intended, the candidate side is about whether anyone told them the tools were there at all. Greenhouse’s 2026 Candidate AI Interview Report, surveying 2,950 active job seekers, found that 63% had already been interviewed by an AI, up 13 percentage points in six months. Only 26% said they trust AI to evaluate them fairly. Thirty eight percent had walked away from a hiring process specifically because it included an AI interview, and another 12% said they would if asked.
The transparency gap is arguably the sharper finding. Seventy percent of candidates in the Greenhouse survey were never clearly told upfront that AI would be evaluating them, and one in five only discovered it once the interview had started. Sharawn Tipton, Greenhouse’s chief people officer, named the failure directly: candidates want employers to “tell them when AI is in the room and what it’s measuring. Right now, most employers are failing that test.”
That is a lower bar than fixing the accuracy of the screening model, and employers are still missing it. Disclosure is a policy decision, not a machine-learning problem, which makes the 70% non-disclosure figure harder to excuse as a rollout growing pain. It reads instead as a choice that most employers, so far, have not had to reckon with because so few candidates have anywhere to escalate the complaint.
One nuance in the Greenhouse data cuts against the easiest version of this story. The report also found that candidates’ perceived rates of bias from AI interviewers roughly matched their perceived rates of bias from human interviewers, meaning candidates are not describing AI hiring as uniquely unfair on the merits. What they are describing is a process that feels opaque regardless of who, or what, is on the other side of it. Only 21% of candidates believe most employers use AI responsibly in hiring at all. The complaint is less “the machine is biased against me” and more “nobody will tell me who or what is deciding, or on what basis,” which is precisely the kind of gap a disclosure policy, not a new model, is built to close.
The Case for AI Screening Has Not Collapsed
None of this data argues that AI hiring tools should be switched off. Employers adopted them for real reasons: high-volume roles generate more applications than any recruiting team can read individually, and a consistent automated first pass can, in principle, apply the same criteria to every candidate in a way a tired hiring manager on their fortieth resume of the day cannot. That efficiency case is intact. What these four sources argue, collectively, is narrower and harder to dismiss: the efficiency case does not automatically deliver a fair or accurate result, and right now the evidence says it frequently is not delivering one, not because AI screening is inherently worse than human screening, but because the specification, the taxonomy, and the disclosure around it have not caught up with how fast the tools were deployed.
What It Means for the HR Leader
None of these four organizations, ACCA, City & Guilds, Andela, or Greenhouse, is arguing that AI has no place in hiring. Read together, they are converging on a narrower and more useful claim: the rollout has outpaced the specification, disclosure, and governance work that would make the tools trustworthy to the people running them and the people subject to them.
For an HR leader evaluating or already running AI screening tools, three implications follow directly from this data:
Audit the taxonomy before the algorithm
Andela’s finding that over half of AI-role job postings mismatch the skills they list suggests the first fix is upstream of any model: rewrite the job descriptions and skill taxonomies feeding the screen before assuming the screen itself is broken. A recruiting team that has not recently reconciled its job architecture against what the role actually requires is asking an AI tool to optimize against a bad target.
Put board-level skepticism to work, not silence it
ACCA’s finding that boards doubt AI hiring more than the workforce average is usually framed as a rollout obstacle. It can instead function as a governance asset: a skeptical board is a natural sponsor for the kind of bias testing, disclosure policy, and human-review-of-AI-decisions framework that regulators (including the EEOC, which HRTech has covered pressing on AI-adjacent bias enforcement) are increasingly expecting employers to have in place before, not after, a complaint.
Disclosure is the cheapest fix on this list
Telling a candidate that an AI system is part of the process, and what it is scoring, costs nothing in engineering time and directly addresses the 70% non-disclosure figure Greenhouse measured. HRTech has previously reported on how job seekers are already outpacing employer AI training on the candidate side of this same gap; disclosure is the one lever employers can pull immediately, without waiting on a vendor roadmap or a new law.
The Gap Closes From Specification, Not Sentiment
The instinct inside many HR functions has been to treat AI hiring skepticism as a perception problem: reassure candidates, reassure the board, and the numbers will follow. The data reviewed here argues the opposite. The distrust tracks specific, fixable defects: mismatched job taxonomies feeding the screen, screening criteria that filter on the wrong skills, and non-disclosure that leaves candidates unable to tell when a decision affecting their livelihood was made by a model instead of a person. Close those three gaps and the trust numbers move because the underlying system changed, not because the messaging did. Leave them open, and every fresh survey between now and next year’s Global Talent Trends report will likely find the same widening gap this one did.
Source: City & Guilds
