Industry Standard
The same AI tools — and the data layers beneath them — are deployed across hundreds of employers. Most risk evaluations can't see what that does to workers.
When HR leaders think about AI risk today, the picture is reasonably consistent. There’s a tool. It scores a candidate, ranks an applicant, predicts a performance issue, or flags a retention concern. Someone is responsible for checking that the tool is accurate, fair, and explainable within the organization’s use. If a worker is harmed, there’s a decision to trace back and, in principle, contest.
That picture is reasonable. It’s also incomplete in a specific way that’s costing real workers real outcomes.
A growing share of the AI shaping people’s working lives doesn’t fit the one-tool-one-company frame. Two layers of concentration sit underneath the visible AI tools in HR. The first is the tools themselves — resume screeners, candidate matchers, video interview scorers — bought by hundreds or thousands of employers and deployed against the same labor market simultaneously. The second is even further upstream: the databases and data products those tools depend on, sold by a smaller handful of providers and licensed across the entire HR tech industry, including to vendors that present themselves as competitors. A worker’s profile travels through both layers — and the same underlying logic — at every employer that buys into either.
This is the aggregate problem. It’s mostly absent from the current AI risk conversation in HR. And it produces real consequences for workers that nobody — not the data provider, not the model’s developers, not the buyer, not the company’s responsible AI program — is positioned to see.
When one logic gates a labor market
The first layer of concentration is the visible one. Many AI tools widely used in HR — resume screeners, video interview scorers, skills inference engines, background check models, candidate-matching algorithms — aren’t deployed by one company. They’re deployed by thousands. The same underlying logic gates access to a huge share of the available employment in any given labor market.
The second layer is mostly invisible to buyers. The data products that power those tools — professional and résumé databases, skills ontologies, compensation benchmarks, public records aggregators, people-graph and identity-resolution platforms — are sold by a smaller number of providers, often licensed simultaneously to multiple competing HR tech vendors. The candidate you’re screening through one vendor’s interview AI may have been profiled, scored, or filtered by another vendor’s tool five times before they ever reached you — using the same underlying data, sometimes literally the same inferred classifications. A change to the data layer propagates outward across every product built on top of it.
From any one buyer’s view, the tool is being evaluated against the work the buyer is doing with it. The buyer’s responsible AI program checks fairness within their use. Compliance reviews the deployment. Procurement vets the vendor. All of it is reasonable work. None of it captures the worker’s experience, and almost none of it reaches the data layer underneath.
The worker isn’t encountering the tool once. They’re encountering it — or a near-clone, or another tool drawing from the same underlying data — every time they apply somewhere. A candidate filtered out by one widely used screener is statistically likely to be filtered out by every other employer using the same product, and very likely to be deprioritized even by competing products that pull from the same data sources. A worker flagged as low-fit by one video interview model is probably flagged the same way across the market. A pay range generated by a dominant compensation benchmarking dataset becomes, by sheer market share, the ceiling and floor for entire labor segments — whether the buyer licenses the data directly or buys an HR analytics tool that does.
The buyer’s evaluation says the tool is operating within tolerances. The worker is hearing the same no, in different rooms, generated by the same underlying logic. The harm is structurally invisible to every individual buyer’s risk evaluation — because the unit at which it becomes visible is the labor market, and nobody is responsible for that.
Why this hasn’t been the conversation
Four reasons.
No single deployment shows the effect. Current risk classifications focus on whether one system, in one deployment, produces fair outcomes within that environment. Most widely deployed tools technically do — they’re checked for bias, accuracy, and explainability inside the buyer’s use of them. The cumulative effect across thousands of deployments is not what any single deployment evaluates. The effect of a third-party data layer that long predates the deployment is even further outside the frame.
It’s evaluated for the wrong audience. Vendors optimize their tools against buyer-side fairness criteria — what looks defensible to an employer’s procurement, compliance, or responsible AI function. The question of what happens to a candidate encountering the same model across forty different employers in a job search isn’t on the evaluation. Neither is the question of what happens to a candidate whose data sits in an upstream database that powers half the tools in the industry. Those questions live in labor economics, I-O Psychology, and measurement science — usually not in the room when the vendor’s pitch is reviewed.
The boundaries are wrong for the problem. Each employer evaluates only what they can see from their own seat. Most procurement reviews evaluate the tool itself; few evaluate the third-party data feeds that power it, even when those feeds are the durable source of any patterns the tool produces. The function with standing to identify the market-level pattern doesn’t exist at any individual organization, and the function with standing to evaluate the data layer underneath doesn’t usually exist either.
It’s hard to regulate and easy to deny. The chain from a single deployment to a market-level outcome runs through many separate buyers, many separate vendors, and a shared data layer beneath them. For regulators, that makes the field genuinely difficult to write rules in — no single actor is responsible for the cumulative effect, and the cumulative effect isn’t attributable to any single actor. For employers, the same difficulty works as a defense. The tool was fair in our use of it. The market effect emerged from many deployments, not this one. The data came from a third party, not us. Each link has someone or something else to point to. Hard to regulate, easy to deny — and that combination is one of the most durable reasons the conversation has stalled.
Why it should matter to you
Even without regulation, the risk is real and it lands in real places.
Operationally. A widely deployed tool is optimized for what works across the average customer — not your specific applicant pool, role mix, or geography. The same model that performs well in the vendor’s benchmark can underperform against your particular workforce, and nothing in the vendor’s evaluation is set up to flag the difference. The same is true of the data layer: a dataset optimized for breadth across the market can be systematically thin or biased against the subset of workers your organization actually wants to find.
Legally. A 2026 class action against Eightfold AI alleges the company built hidden “consumer reports” on job applicants — assigning each a 0-to-5 score based on scraped personal data, and screening out low-scoring applicants before any employer reviewed them — without the disclosures required under the Fair Credit Reporting Act and California consumer protection law. The legal theory isn’t discrimination. It’s that an AI screening tool deployed across many employers, drawing on data the applicants never consented to, can trigger consumer protection statutes no single employer thought applied. If the tool your organization relies on is shared across your industry — or if its underlying data layer is — the next legal angle on it may come from a statute nobody was watching. “Industry standard” is not a defense when the standard itself becomes the target.
Reputationally. The patterns become visible eventually. Workers compare notes. Journalists trace them. The organizations that examined how their shared vendors and shared data sources were affecting the broader labor market before they were forced to will tell a better story than the ones that didn’t.
Strategically. An HR function relying on the same tools as its peers is competing for talent through the same filters as its peers. The candidates those filters discard are discarded across the market. The candidates they surface are the ones every competitor is also chasing. Even an organization that has differentiated its HR tech stack is usually drawing from the same underlying data — making the differentiation more cosmetic than real. The strategic value of any tool decreases as the data layer beneath it converges. The shared risk does not.
The risk of doing nothing isn’t that something dramatic happens tomorrow. It’s that the slow accumulation of consistent outcomes — for real workers, generated by the same underlying logic across many of your peers, drawn from the same upstream sources — becomes the basis of how the labor market works. By the time anyone names the problem, it’s structural.
Where to start
Bringing this face of the aggregate problem into the conversation doesn’t require new technology. It requires asking different questions. Four are usually enough to surface the gap.
Which AI tools in our HR portfolio are also deployed by a substantial share of our competitors, peers, or the broader employer base in the same labor market?
Which third-party data sources, databases, or upstream providers feed those tools — and do any of our competitors’ tools depend on the same sources?
Has anyone evaluated the cumulative effect those tools and the data layer beneath them may produce at the labor market level — beyond their evaluated effect within our own use?
Where does our responsibility end — at the boundary of our own deployment, or at the boundary of the workers our deployment affects?
If those questions are uncomfortable, the aggregate AI being used across our industry is producing real consequences for real workers under evaluation frameworks that don’t see them. The first step is to look at the right scale. Better to do it now than after something else forces the issue.
Work Science Consulting LLC (WSC) provides independent, science-based advisory to organizations navigating AI in HR and workforce contexts. worksciconsulting.com