Artificial intelligence is being adopted into HR processes at a pace that most HR practitioners are not adequately prepared for, and the gap between the pace of adoption and the depth of the profession’s understanding of its implications is a genuine and growing risk. The risk is not primarily that AI will be adopted without benefit, though that is certainly possible and in some cases is already occurring. The primary risk is that AI will be adopted without adequate attention to the specific dimensions of AI deployment in HR contexts that are most consequential for the people whose employment decisions are shaped by these systems.
The HR practitioner who positions AI as primarily a technology question, to be evaluated primarily by the IT and data science functions with HR input on the people processes to which the AI will be applied, is missing the specific expertise that HR should be bringing to the evaluation and governance of AI in employment contexts. AI in HR is fundamentally a fairness, ethics, and human dignity question that has a technology implementation dimension, rather than a technology question that has a people dimension. This positioning inversion determines who leads the conversation, what questions are asked, and what safeguards are built into the deployment. Getting it wrong has specific, documented, and consequential effects on the people whose employment decisions the AI is shaping.
What HR AI is actually doing in most organisations
The adoption of AI into HR processes is already substantially advanced in most large organisations, though the visibility of that adoption varies considerably. Recruitment process automation, including AI-powered CV screening, candidate matching against job specifications, chatbot-mediated initial screening, and video interview analysis, is now widespread in large organisations and is influencing the employment decisions of millions of candidates in ways that most of those candidates are unaware of. Performance management analytics, including the use of employee activity and output data to generate performance assessments, is growing in adoption particularly in sales, customer service, and knowledge work contexts where digital activity generates sufficient data for automated analysis. Workforce planning and predictive attrition models, which use historical employment data to identify employees at risk of leaving and to optimise headcount decisions, are being deployed in an increasing number of large organisations.
Each of these applications involves the use of historical employment data to make or inform decisions about current employees and candidates. The specific risk this creates, which is well-documented in the academic and policy literature even if it is insufficiently attended to in practitioner discussions, is the perpetuation and amplification of historical bias. AI systems that are trained on historical employment data will, absent specific intervention, learn and reproduce the patterns in that data, including the patterns that reflect historical discrimination, historical exclusion, and the specific features of the historical environment that produced employment outcomes that the organisation would not now endorse as fair or appropriate. The AI system that learns from historical promotion data in which women were promoted at lower rates than men, and that uses that learning to score future promotion candidates, is not merely reflecting historical patterns. It is encoding those patterns into automated decision support in ways that make them both more efficient and more difficult to detect and challenge.
The specific governance questions that HR must own
The governance of AI in HR is a function that HR must own rather than delegate, because the specific questions that responsible AI governance in employment contexts requires are questions that sit within HR’s expertise and responsibility. They are not primarily technical questions, though they have technical dimensions. They are questions about fairness, transparency, accountability, and the protection of employee and candidate rights in the context of automated decision-making.
The first governance question is transparency: do the people whose employment decisions are being shaped by AI systems know that AI is being used, and do they have the information they need to understand how it is being used? This is both an ethical requirement and, in many jurisdictions, an emerging legal one. The candidate whose CV is being screened by an AI algorithm without their knowledge is in a qualitatively different situation from the candidate who knows this and can adjust their application accordingly. The employee whose performance is being assessed through digital activity monitoring without their knowledge is in a situation that raises specific ethical and legal concerns that HR must address rather than delegate.
The second governance question is fairness: has the AI system been assessed for the specific demographic patterns in its outputs, and have those patterns been evaluated against the fairness standards the organisation applies to human decision-makers? This assessment requires both the technical capability to analyse the outputs of AI systems for demographic disparities and the analytical understanding to distinguish between disparities that reflect genuine performance differences and those that reflect the encoding of historical bias in the training data or the algorithm design.
The third governance question is accountability: when an AI-informed employment decision produces an outcome that an employee or candidate believes was unfair, what is the process for reviewing that decision, who is accountable for the outcome, and what is the employee’s or candidate’s right to challenge it? The AI system does not have accountability in the meaningful sense. The people and the organisation that designed it, deployed it, and relied on its outputs for employment decisions do. HR must design the accountability structure that ensures that the human accountability for AI-informed employment decisions is clearly assigned and genuinely exercised.
Building the HR capability to govern AI responsibly
The specific capability that HR needs to govern AI in employment contexts responsibly is a combination of conceptual literacy and practical skills that most HR functions have not yet developed systematically. The conceptual literacy includes a working understanding of how machine learning systems work, what training data is and how it shapes algorithmic outputs, what algorithmic bias is and how it manifests in employment contexts, and what the specific regulatory and ethical frameworks that govern AI in employment decision-making require. The practical skills include the ability to evaluate AI vendors’ claims about their systems’ fairness and transparency, to design the auditing processes that assess AI systems’ outputs for demographic disparities, and to build the governance frameworks that ensure human accountability for AI-informed employment decisions.
This capability is not primarily a technical capability. It is a capability in the governance and ethics of human capital decision-making in an environment where those decisions are increasingly shaped by automated systems whose logic is not always transparent and whose fairness cannot be assumed. Building it should be among the most urgent priorities for HR functions that are already deploying AI in their employment processes or that are planning to do so.
The regulatory environment HR must anticipate
The regulatory environment for AI in employment contexts is evolving rapidly in ways that HR functions must anticipate rather than react to. The European Union’s AI Act classifies several HR AI applications, including CV screening systems, performance assessment systems, and behaviour monitoring tools, as high-risk AI systems subject to specific requirements for transparency, human oversight, and pre-deployment conformity assessment. Beyond the EU AI Act, multiple other jurisdictions are developing regulatory frameworks for AI in employment contexts. The HR function that is not actively monitoring this regulatory environment, and that is not building the governance infrastructure to achieve compliance with applicable requirements, is creating the kind of legal and reputational exposure that organisational governance requires to be managed proactively rather than reactively. The compliance investment that responsible AI governance in HR requires is considerably less expensive than the remediation required when non-compliant AI deployment produces the documented discriminatory outcomes that current enforcement actions are beginning to address.
The HR functions that are most effectively governing AI in employment contexts share one distinctive feature: they have invested in building genuine understanding of the AI systems they are deploying, rather than accepting vendor claims about system fairness and transparency at face value. This understanding requires the combination of technical literacy and critical evaluation skills that most HR functions are only beginning to develop. The investment required to develop it is substantial but clearly warranted by the specific legal, ethical, and commercial risks that AI deployment in employment contexts creates. The HR function that treats AI governance as a technical function to be delegated to IT and data science, while retaining oversight in name only, is not providing the quality of human capital governance that responsible AI deployment in employment contexts requires.
The adoption of AI in HR is not primarily a technology question. It is a fairness question with technology implications. The HR function that governs AI deployment in employment contexts with the same rigour and the same ethical responsibility that it applies to human decision-makers is providing the most important safeguard available against the specific risks that automated decision-making creates for the people whose employment decisions it shapes.