In 2024, Derek Mobley, an African-American man over 40, living with a disability, applied to more than 80 jobs that used Workday's AI-powered screening tools. He was rejected every time. His subsequent lawsuit alleged discrimination on the grounds of race, age, and disability.
In May 2025, a federal court in the Northern District of California conditionally certified the case as a nationwide class action. Workday disclosed in court filings that its software tools had rejected 1.1 billion applications during the relevant period.
That number is not hypothetical. It is now being examined in a federal courtroom. And for anyone involved in hiring, it should be impossible to ignore.
The Bias Problem is Structural, Not Algorithmic
The instinctive response to cases like Mobley v. Workday is to blame the algorithm: fix the model, run bias audits, and add human oversight. Mrityunjaya P, Chief Technology Officer at MAI Labs, argues this framing misses the deeper issue entirely.
“The problem is not that AI hiring tools are poorly designed. Many of them are technically sophisticated. The problem is what they are being asked to do: make decisions based on resumes,” he tells AIM.
Resumes, he contends, are the least trustworthy document in the hiring process. Studies show around 17% of resumes in India contain significant discrepancies, with similar patterns observed in the US. Job titles are inflated, responsibilities exaggerated, and credentials sometimes claimed from institutions that either do not exist or never granted them.
When AI is trained on such data, it learns to pattern-match against whatever correlated with historical “successful” hires. Because historical hiring was itself biased—favouring certain schools, companies, and demographics—the AI reproduces those biases at scale.
Research supports this. In 2025, Stanford researchers found that AI resume-screening tools rated older male candidates more highly than female and younger candidates, even when all resumes were generated from identical underlying data. Research by VoxDex found that AI tools systematically favoured female applicants over Black male applicants with equivalent qualifications.
The algorithm, then, is not the issue. The problem lies in the input data, unverified resumes and historically biased hiring patterns.
The issue is not confined to the US. In India, highly experienced and skilled professionals are also being rejected during AI-led initial screenings.
“Company AI systems reject CVs, while candidate AI tools fight to break through. It becomes a battle of algorithms, with humans left on the sidelines. The CV that makes it through is not necessarily from the best person for the job, it’s the one that best managed to trick the algorithm,” says Ritesh Dalal, CEO of Intellective. “Meanwhile, brilliant candidates who don’t know how to optimise for AI screening are filtered out before a human ever sees their application. In effect, we’ve automated ourselves into a worse outcome than we began with.”
Metrics That Mask, Not Measure
The bias crisis in AI hiring does not exist in isolation but it sits atop a broader failure in how the industry has measured recruitment itself. According to Abhishek Agarwal, President at Judge India and Global Delivery at The Judge Group, hiring functions have historically been managed by activity rather than impact.
“Somewhere between posting a job description and a new hire’s first performance review, most organisations lose the thread. They track the steps, applications received, interviews scheduled, and offers made, without ever connecting those inputs to the outcomes they were supposed to produce,” says Agarwal.
That model is now being challenged from two directions: regulatory scrutiny demanding accountability over outcomes, and AI itself providing tools to measure them more effectively.
New KPIs are gaining traction. Time-to-value—the period from offer acceptance to meaningful productivity—is replacing time-to-hire. Quality-of-hire composite scores, based on performance data and structured manager feedback at 30, 90, and 180 days, are replacing gut instinct. Real-time funnel analytics are displacing lagging weekly reports.
“A fast hire that underperforms for 18 months before leaving costs more, in every measurable way, than a longer, more deliberate process that brings in someone who delivers and stays,” Agarwal adds.
The Real Fix: Better Data, Not Better Algorithms
For Mrityunjaya, the path forward is not more sophisticated models but changing what those models rely on. Algorithmic audits and disparate impact testing are necessary, but insufficient, if the underlying input remains an unverified, self-reported document.
"If the AI is screening resumes, and the resumes are unverified, the AI is making decisions based on claims it has no way to confirm. No amount of algorithmic auditing fixes that,” Mrityunjaya says.
He points to verified credential infrastructure, blockchain-anchored records, and AI-assessed skill scores as the structural answer. When degrees are cryptographically verified rather than self-reported, and skills are assessed rather than claimed, AI is no longer pattern-matching against noise. It is evaluating evidence.
This approach removes bias from unreliable input data—the largest documented source of AI hiring bias—at its source, rather than attempting to patch it downstream.
The implications are particularly significant for India, where AI adoption in hiring is accelerating across IT services firms, startups, and recruitment agencies–-without a corresponding regulatory framework for fairness. Given India’s high rate of resume discrepancies and the lack of verified credential systems, the conditions that produced Mobley v. Workday already exist and remain largely unaddressed.
A Reckoning That is Already Arriving
Regulatory pressure is building. Four US states now have AI hiring bias laws in place. The EU's AI Act is approaching implementation. Hundreds of discrimination complaints have already been filed against AI hiring tools.
Agarwal notes that metrics such as revenue per recruiter and cost-per-quality-hire are beginning to appear on executive dashboards, signalling a shift in how talent acquisition is perceived.
"The organisations building the capability to do it are pulling ahead. The ones still running on spreadsheets and intuition are starting to feel the gap,” Agarwal adds.
What emerges from both perspectives is that AI has not made hiring more objective. It has made hiring faster, and in doing so, has scaled whatever biases were already present in the data and the decisions that trained it.
Your reaction
Discussion
What's your take on this story?
No comments yet — be the first to share your view.