When several employers use the same hiring vendor, a candidate may be assessed by the same system more than once. Each employer may still make its own decision. But the model, assessment method or even some of the candidate's data may be the same.
The problem is that candidates usually cannot see which of those is happening.
As conversational hiring agents spread through Southeast Asia and markets such as South Korea, that raises a question: can a candidate, an HR team or a regulator tell whether one vendor's recommendation is influencing decisions at more than one employer?
One rejection, many employers
A recent US study provides a useful starting point.
On 26 May 2026, Rishi Bommasani, Sarah H. Bana, Kathleen A. Creel, Dan Jurafsky and Percy Liang published Algorithmic Monocultures in Hiring, later presented at the 2026 ACM Conference on Fairness, Accountability and Transparency. They analysed 4,197,168 applications from 3,372,132 applicants applying to 1,746 positions at 156 US employers. Every application in the dataset was screened by pymetrics.
On average, 41.8% of applications received a “do not recommend” outcome. Because these were high-volume hiring processes, the researchers found that a hiring manager was unlikely to consider a candidate further after that result.
The more striking finding came when people applied repeatedly. Among applicants who applied to four positions, 10% were recommended for rejection from all four. Among those who applied to 10 positions, 4% were recommended for rejection from all 10, a higher rate than independent decisions would predict. The researchers identified 42 pymetrics models being used at more than one company.
The candidate data matters too. Pymetrics stored features generated from its assessment games and reused them when the same person applied for another pymetrics-mediated job within 330 days. The candidate was not necessarily starting again with a fresh assessment. The same underlying features could be scored by another model. That 330-day reuse is specific to the US pymetrics dataset. There is no equivalent published retention period in the APAC evidence examined here.
The researchers estimate that applicants would need about 25 applications to have a 99.9% chance of receiving at least one recommendation, compared with about 10 if each decision were independent. The same assessment infrastructure can shape outcomes across multiple applications.
That is the part worth watching as these systems spread across APAC.
The tools are arriving in APAC
Talent teams in Southeast Asia are already experimenting with AI in recruitment, although the public evidence tells us much less about which vendors sit behind those systems.
LinkedIn's Future of Recruiting 2025 South East Asia edition, based on a September 2024 survey of 1,271 recruiting professionals across 23 countries, found that 79% of Southeast Asian talent-acquisition professionals agreed AI would change how organisations hire. Among Southeast Asian organisations, 29% were actively integrating or experimenting with generative AI in hiring, compared with 37% globally.
South Korea provides another signal. A JobKorea survey of 1,286 corporate recruitment officials found that about 65% were considering or using AI recruiting agents. The survey does not identify which systems are deployed.
Singapore has a higher reported level of AI use, although the definition is broader. HR Online, citing ManpowerGroup's Employment Outlook on 11 March 2026, reported that 82% of 538 Singapore organisations were using AI in hiring, onboarding or training. The figure was 81% across Asia Pacific and the Middle East and 67% globally. Learning and development was the most commonly cited use at 32%, rather than hiring.
An India-based vendor study by Taggd, reported by ANI in August, found that nearly 80% of organisations were using AI in at least one recruitment function, while only 10% reported measurable improvements in hiring outcomes.
The numbers are not directly comparable. They use different definitions and populations. But they point in the same direction: AI is moving into recruitment faster than the public record is documenting how these systems work.
The same agent, two employers
Let's look at two Singapore employers which have publicly identified the same recruitment technology provider.
Singtel uses SIEA, a conversational AI assistant powered by Workday Paradox. Workday says SIEA helps candidates find jobs, applies screening checks and handles interview scheduling.
KPMG Singapore uses Kiara, another Paradox recruitment assistant. Its publicly described functions include answering candidate questions and scheduling interviews.
This establishes that the same vendor can sit inside the recruitment process at more than one employer. It does not establish that the employers share candidate scores or assessment data.
Singtel's customer story describes a group of about 25,000 employees across Asia Pacific making about 4,000 hires a year. Joshua Teo, head of group strategic resourcing and talent acquisition, is quoted on SIEA. Workday attributes a 75% reduction in time-to-hire, 5,000 administrative hours saved and a fall in interview rescheduling from 19% to 10%. These are company and vendor-reported results.
KPMG partnered with Paradox in 2025 to deploy Kiara. HRM Asia described the agent as handling candidate questions and interview scheduling.
The public descriptions therefore show a shared vendor, but not a shared candidate record. Neither customer story says how recommendations are formed, how long candidate information is retained, or whether another Paradox customer can reuse it.
That is the gap between what the market can see and what a candidate can know.
What you cannot see
Recruitment vendors sell speed and scale. What's worth considering is what happens inside the screening process and what information remains attached to the candidate afterwards.
Singapore's own employment guidance recognises the problem. TAFEP says AI can make it difficult for employers and HR teams to understand how a score or recommendation was formed, and warns that poorly designed systems can repeat bias or filter out qualified candidates with non-traditional profiles.
The issue is already being tested in the United States. In June 2026, CNA reported, citing Reuters, that a US judge allowed a class action against Workday to proceed. The plaintiffs allege that Workday's screening tools helped exclude applicants across employers. Workday has said its tools do not make hiring decisions.
That remains a US dispute, and it puts a concrete legal question around a problem that can otherwise remain invisible: when a vendor sits between candidates and several employers, where does responsibility for the screening decision sit?
In 2023, the US Equal Employment Opportunity Commission reached a USD 365,000 settlement with iTutorGroup after software automatically rejected US-based women aged 55 and over and men aged 60 and over. More than 200 qualified US-based applicants were affected.
That settlement shows an automated hiring system can fail in production. A shared-vendor system producing the same effect in Singapore, South Korea or elsewhere in Southeast Asia remains unmeasured.
Who still owns the hire
Singapore has been clear about where accountability sits.
In October 2025, TAFEP published Fair Hiring First, AI Second, stating that employers, not algorithms, remain accountable for fair, transparent and compliant hiring decisions. Its guidance says employers using AI should use job-related data, keep people involved in key stages, retain records and give candidates a way to ask questions or appeal.
The Workplace Fairness Act has been enacted but is not yet in force. Commencement remains expected rather than fixed.
That creates an interesting practical problem. The employer remains accountable for the decision, but some of the information needed to reconstruct an automated screening decision may sit with the vendor.
A 24-hour scheduling agent can still be useful. It takes repetitive work off a recruiter's calendar. But the responsibility for the hire does not move with the work.
What this boils down to is whether an employer can explain what the system did, what information it used and whether the same system is influencing decisions elsewhere.
For candidates, that could matter as much as the final hiring decision itself.
What to watch
Which employers share a screening vendor. Two Singapore employers using Paradox are visible because the companies or vendor published customer stories. Most candidates will not know which system is assessing them.
Whether candidate assessments travel. The US pymetrics study showed that assessment features could be reused across applications within 330 days. No equivalent APAC retention or reuse practice has been publicly established.
What happens when the Workplace Fairness Act takes effect. Commencement remains expected rather than fixed. How employers, vendors and candidates handle AI-assisted decisions under the new framework will provide a more concrete test of accountability.
