Stanford University released the first large scale study of hiring algorithms in May 2026, which asks whether there is an impact on job seekers when employers all use the same algorithm for hiring. Ninety percent of US employers use AI screening tools during hiring, relying on a small selection of third party vendors to do so. When hiring algorithms are used for application screening by multiple employers an algorithmic monoculture is created. The Stanford Researchers say that algorithmic monocultures occur when decision-makers rely on the same or similar algorithms. They set out to understand whether, in such circumstances, the algorithms would result in the same groups of applicants being rejected across separate organisations.
The paper follows 3.4 million people who submitted 4 million job applications across 150 employers and 11 industry sectors in the US. The research team took a look inside the ‘black box’ of algorithmic hiring to determine what impact AI is having on the new workforce. They found that AI tools increase racial bias and systemically shut out the same candidate out of jobs, irrespective of which role they apply for. In total, 26% of black applicants and 15% of asian applicants were subject to systemic AI discrimination against their racial group.
A shared dependence on a single hiring vendor across multiple independent recruiting systems creates homogenous outcomes. Someone who submits multiple applications screened by the same algorithmic hiring vendor is more likely to be rejected from all of them. This risk of rejection is higher than it would be if each company made its hiring decision independently. When a single hiring vendor dominates screening for an industry, rejection practices become systemic. There are a small number of hiring vendors across all sectors, and single vendors have significant capture in certain industries. Therefore, the issues in their screening processes amplify the consequences of biased and discriminatory outcomes, causing the racial bias to become systemic in sectors.
If you only rely on the numbers, then every candidate receives some type of offer. But the big picture obscures how selective criteria used by AI tools is being justified by race. This means that without digging into the details, adopting AI hiring tools makes it look like every candidate is being fairly judged for a role. In reality, there is a hierarchy of which roles are being recommended to certain types of people.
Researchers at Stanford conclude that, “AI screening tools bring together three properties that should not co-exist in high-stakes decision-making: they are pervasively adopted, highly consequential, and opaque to the public.” They call for more independent research into algorithmic hiring to illuminate the consequences of these tools and inform evidence-based AI policy.
The report’s four recommendations aim to make algorithmic hiring accountable not only within an employer, but across the labour market when many employers use the same vendor.
1. Test adverse impact per position
The study found that especially for black and asian applicants, aggregated figures produced from AI hiring systems can appear fair, but in reality they conceal substantial disparities in particular roles. Employers should require vendors to provide demographic selection-rate and impact-ratio results for every model or role used.
2. Strengthen market surveillance
Public agencies should be able to observe how hiring-algorithm vendors operate across their client base, rather than relying on investigating employers in isolation. The rationale is that a vendor can influence screening decisions across many organisations, so interrupting this system can reduce the harms and patterns of exclusion that become distributed across the market and invisible in any single firm’s data.
3. Monitor algorithmic monoculture
We should consider dependence on common vendors, models and assessment methods as a systemic labour market risk. Agencies should begin to track dependency to understand how patterns of rejection could be correlated.
4. Mandate independent researcher access
Legislators should create lawful routes for independent researchers to access appropriately protected algorithmic-hiring data, models, and outcomes. The authors argue that independent access is needed because vendor claims and employer-level audits cannot, on their own, reveal cross-employer effects such as systemic rejection; their own study depended on unusual access to data covering roughly 3.37 million applicants and 4.2 million applications.
For a DEI or workforce-governance team, the practical message is: do not accept a supplier’s overall “bias-free” claim as sufficient. Contract for role-level impact reporting, model/version traceability, and information about how widely the vendor’s screening approach is used.
The Stanford research proves that algorithmic monocultures in hiring are having structural shifts to the labour market in the US. Businesses based in the UK that are also engaging algorithmic hiring systems will likewise be vulnerable to creating and sustaining algorithmic monocultures within the British business context. Algorithms have influence across individual employer level and across sectors, where organisations are reliant on the same hiring systems. As such, algorithmic monocultures are exacerbating preexisting exclusion of select demographics, including black candidates. A hiring vendor may be used by a recruiting system with a generalised understanding that it offers ‘fair’ outcomes, but the Stanford research proves this is not always the case. In particular, black and asian applicants are more likely to be systemically rejected than other candidates, when a single hiring vendor is in place. AI systems are being used at a huge scale and are to a large extent, unregulated. Stanford researchers call for more regulation from auditors and for hiring systems to take more care to investigate the claim to be ‘fair’. There is a possibility that this outcome may decrease once more AI hiring vendors are adopted into the hiring process. Conversely, the opposite may just as well be true. Whilst hiring workforces and applicants should be made aware of this discriminatory outcome, as a first step, it remains clear that the impacts of hiring algorithms at scale is insufficiently understood.
To find more detail, read the paper here: https://hai.stanford.edu/news/ai-hiring-tools-can-yield-racial-bias-and-systemic-rejection
Written by
