AI in Hiring Interviews: High Trust, Unresolved Bias in HR

The paradox of declared trust

Two figures published in the BambooHR report of September 2, 2026, conducted with over 500 HR professionals, define the fracture that AI governance in hiring must address today: 67% report trust in AI to conduct interviews autonomously, while 80% of the same respondents have observed bias or issues in the language models supporting that work, as reported by HCAMag^[1].

The gap between these two figures is the real governance question for 2026.

Those who report trust and those who observe bias are, largely, the same people: professionals who recognize the problem and, simultaneously, authorize the tool to decide who enters an organization.

Demographic bias and model distortion: what the numbers show

The report breaks down the phenomenon into measurable clusters. 36% of HR professionals have observed AI favoring resumes and cover letters generated by the same model evaluating them, a circular pattern that rewards linguistic form, never candidate merit. 33% report recurring demographic bias patterns in their selection tools, according to the same report.

Concurrently, three out of four HR professionals (76%) describe themselves as daily, confident users of AI, compared to just 50% recorded among general workers. The trust gap between those who govern the tool and those who experience its effects grows, never shrinks, as adoption advances.

Mirroring: when humans copy the machine's error

A University of Washington study, cited in the BambooHR report, offers evidence of the mechanism that makes this gap dangerous. Researchers observed that human decision-makers, faced with distorted AI output, tend to mirror the machine's decision instead of correcting it.

The phenomenon reverses the common assumption that human oversight functions as an automatic corrective for algorithmic bias. A reviewer who simply confirms AI output adds an appearance of control, never actual control.

An organizational design failure, not a technology one

The correct reading of these numbers concerns organizational design, not technology itself. An organization that promotes AI to the role of interviewer, knowing that 80% of its professionals have observed its biases, is delegating a human capital decision to a system it itself recognizes as partial.

Nicole Csizar, senior director of HR services at BambooHR, frames it this way: AI should make the interviewer better at the part of the work requiring human presence, never replace them entirely, as reported by HCAMag^[1].

92% of organizations are already piloting or planning AI agents across the entire HR cycle: screening, attrition risk monitoring, policy application. The speed of adoption systematically outpaces the speed at which governance infrastructure is built.

What organizations closing the gap are doing

Organizations that narrow this divide share recognizable, industry-documented practices.

• Periodic and independent reviews of output patterns from models used in selection, never just an initial audit at tool purchase
• Clear thresholds beyond which a decision requires substantial human review, never mere pro forma confirmation
• Training reviewers on the mirroring effect documented by the University of Washington, so that oversight remains an active corrective

A 2024 study conducted in Australia had already flagged that HR leaders viewed algorithmic discrimination against underrepresented workforce groups as probable. The issue predates this report by nearly two years and remains unresolved.

What CEOs, CHROs, CFOs and boards need to know

For the CEO, the governance data to bring to the board concerns organizational readiness in AI hiring governance, not solely the scope of adoption. A 92% piloting rate accompanied by absent bias audits describes exposure, not maturity.

For the CHRO, the immediate priority is building a verification process for selection models, on a regular schedule with responsibility assigned to a specific role.

For the CFO, the reasoning is economic risk: a controversy over algorithmic discrimination in hiring costs far more in reputation and litigation than a quarterly audit of model patterns. For the board's talent and compensation committee, the metric to track remains straightforward: the share of AI-assisted hiring decisions subject to substantial, not symbolic, human review.

The design question for leadership

The 67% and 80% describe two largely overlapping populations: professionals who use the tool daily and, simultaneously, know its documented limitations.

The open question for any leadership concerns the verification structure capable of making that trust earned, not the level of abstract trust declared. An organization that answers this question with a process, assigned roles, and recurring audits transforms a risk metric into a documented competitive advantage.