Is AI candidate screening biased?
It can be, and federal courts are testing exactly that. Mobley v. Workday, a collective action alleging AI screening tools filtered out applicants by age, race, and disability, survived key motions to dismiss in 2026. The risk concentrates in opaque resume-stage filters that reject people in minutes without a conversation. Structured interviews pull in the opposite direction, with each candidate answering the same questions, scored against the same rubric, with written reasoning a human can audit.
Where AI hiring bias shows up
The highest-profile test is Mobley v. Workday, a federal case in the Northern District of California alleging that algorithm-based applicant screening discriminated by age, race, and disability. The court ruled in 2024 that a vendor performing screening functions can be treated as the employer's agent under federal anti-discrimination law, conditionally certified a nationwide age-discrimination collective in May 2025, and denied further motions to dismiss in June 2026. Forbes reported that the court-authorized opt-in notice could make it one of the largest collective actions in U.S. history. The full docket is public.
The allegations describe proxy discrimination: screening data that stands in for protected traits, like employment gaps correlating with medical conditions. The earlier facial-analysis era produced its own reckoning. In March 2025, the ACLU filed a complaint on behalf of a Deaf, Indigenous woman whose AI video interview lacked working captions and whose automated feedback told her to "practice active listening."
The pattern across these cases: bias concentrates where a system evaluates something other than what the candidate actually said, and where nobody can inspect why a score was given.
Why structured interviews reduce bias
Decades of industrial-organizational research support structured interviewing as one of the most valid and least biased selection methods available. The mechanism is simple. When every candidate answers the same questions in the same way, mood, small talk, scheduling luck, and affinity effects drop out of the evaluation. When answers are scored against a written rubric instead of gut feel, the criteria are visible and challengeable. Unstructured screening is where interview bias actually lives.
Structure alone is necessary but insufficient. The other half is auditability: a scoring decision nobody can review is a scoring decision nobody can correct.
How Talent Pronto is designed for audit
Talent Pronto applies the structured-interview model and adds the paper trail:
- Every candidate for a role answers the same questions in a text conversation, scored against the employer's rubric. Knockout questions are pass/fail; behavioral answers are scored 0 to 100 with written reasoning for every score.
- The interview is text-based. There is no camera, no facial analysis, no voice biometrics, and no emotion inference. Anna evaluates the content of answers against the rubric.
- No candidate is auto-rejected by the system. Every applicant stays visible with their scores and transcript, and the hiring team makes every decision.
- Every interview produces a complete audit trail: questions, answers, per-criterion scores, reasoning, and access logs showing who viewed the record.
- Candidates who need accommodations have alternative arrangements available, and anyone can opt out of the AI screen with no penalty.
This is the same design that regulated government agencies use for bias-reducing interviews at scale. The full picture is on the security, compliance, and fair hiring page.
Common follow-up questions
Doesn't a rubric just encode the employer's bias?
The rubric makes the criteria explicit, which is the point. Written criteria applied identically to every candidate can be reviewed, challenged, and corrected. Gut-feel judgments spread across dozens of screeners cannot.
What should we ask an AI screening vendor about bias?
Ask what data the system evaluates, whether every candidate gets identical questions and criteria, whether scores come with reviewable reasoning, whether the system can reject anyone without a human, and what records exist if a decision is challenged.
Does the Workday case mean employers are liable for a vendor's tool?
Employers were always liable for discriminatory outcomes in their own hiring. The Mobley rulings signal that vendors may share that liability when their tools perform screening functions. Either way, the employer's exposure makes transparency and human oversight a requirement, whatever tool sits in the funnel.
