AI in HR: A Tool with Promises and Risks

The value of AI in HR is not measured by the number of automated tasks, but by the concrete progress it brings to employees, without sacrificing human trust or responsibility. AI can improve the employee experience, support HR teams, and save managers time. However, it also amplifies existing vulnerabilities. Therefore, its deployment must start from business needs, rely on reliable data, incorporate controls, and demonstrate real benefits.

AI's Growing Presence in Human Resources

Artificial intelligence is gradually becoming established in human resources departments. It promises to respond more quickly to employees, lighten certain administrative tasks, and help managers find a rule or prepare an interview. However, according to the Deloitte 2026 Global Human Capital Trends report, only 6% of executives declare progress in designing interactions between humans and AI (Deloitte, 2026). This figure brings up a more decisive question: what happens when a powerful tool relies on an imperfect process?

AI as an Amplifier of Both Benefits and Risks

The answer lies in a simple idea: AI is an amplifier. It can make HR experiences simpler, more accessible, and more fluid. However, it can also make an error faster, more credible, and harder to detect. The challenge is therefore not to integrate AI everywhere, but to choose the irritants to solve, the guarantees to establish to preserve trust, and the real benefits to demonstrate.

The Role of Data and Controls in AI Deployment

This risk of amplification reminds us that the most costly failures do not always come from defective technology. They often arise from a poorly mastered data point, rule, or interface. The story of the Mars Climate Orbiter has become a case study. In 1999, after nine months of travel, NASA lost this probe as it approached Mars. One team used imperial units, the other metric units. The trajectory calculations gradually diverged, placing the device too low.

Similar Failures in Financial Systems

The financial sector experienced a similar error. In 2005, an operator at Mizuho Securities was supposed to sell a J-Com share for 610,000 yen. The order transmitted was the opposite: 610,000 shares at one yen. The Japanese regulatory authority then noted deficiencies in training and supervision in the processing of orders. The decision by the Japanese Financial Services Agency explains this. These episodes do not concern AI. However, they show what happens when a misinterpreted data point, an ambiguous instruction, or an absent control becomes the starting point of a powerful system.

Risks in HR AI Applications

This is precisely the risk to anticipate in HR uses. Thus, an imprecise definition of "departure risk," incomplete performance data, or poorly constructed salary equity indicators can guide a large number of recommendations. The European directive on pay transparency provides for the monitoring of the pay gap between women and men, including for variable components, median gaps, pay quartiles, and categories of workers. If these indicators rely on fragile data or definitions, AI risks accelerating their spread rather than helping to correct inequalities.

AI as a Tool for Accessibility in HR

This observation should not be an argument against AI. It allows, on the contrary, to better define the situations in which it can provide concrete help, provided it is deployed on solid foundations. For an employee, a conversational interface can simplify access to HR services: understanding a leave procedure, preparing a mobility, finding information on pay, or identifying a training. It can reformulate a request, guide a person step by step, and direct them to the right contact.

Conditions for AI's Success in HR

This function can be particularly useful for people unfamiliar with digital tools or internal organization. However, it only fulfills this promise under two conditions: that the answers are reliable and that a human contact remains reachable when a situation is singular or sensitive. The same principle applies to professionals who support employees. For managers, AI can help prepare an interview, synthesize information, or find applicable rules.

The Importance of Human Oversight and Data Quality

For HR teams, it can produce a first response, format a document, flag inconsistencies to check, or prepare candidate and action recommendations. Again, the quality of the responses and the effectiveness of AI will depend less on the tool itself than on the reliability of the data, the solidity of the management processes, and the clarity of the roles and responsibilities of each person. The time freed up by AI should not serve only to absorb an increase in demands.

Reinvesting Time in Strategic HR Projects

The Deloitte 2026 Global Human Capital Trends study, on the contrary, invites to reinvest it in projects that have become central to the HR function: redesigning the division of work between humans and AI, anticipating the evolution of skills, supporting managers, and developing a culture of continuous learning. HR will also have to define and maintain the essential conditions of trust: data quality, transparency of uses, clarity of responsibilities, and preservation of equity.

Regulating AI to Maintain Trust

For AI to provide effective help without undermining trust, its use must be regulated. The organization must define what the system can do, the decisions it cannot make alone, and the way in which a person can contest or correct its responses. Verifying responses that commit an AI can produce a smooth, apparently precise response, but inaccurate. In 2023, American lawyers submitted to a court non-existent court decisions generated by an AI. The references appeared sufficiently plausible to not be verified before their submission.

Legal Consequences of AI Misuse

The federal court in New York sanctioned the lawyers and their firm in the case Mata v. Avianca. When a response concerns a right, pay, career, recruitment, or any other individual situation, AI must assist the decision, not replace it. Protecting data and rights Beyond the reliability of responses, the nature of the information processed requires particular vigilance. HR systems concentrate particularly sensitive information: professional careers, salaries, evaluations, absences, even health data or data related to diversity.

Legal and Ethical Considerations for AI in HR

In France, the CNIL reminds that AI systems processing personal data must pursue a defined, explicit, and legitimate purpose, and limit the data used to what is necessary. Uses related to recruitment, evaluation, or promotion also require particular vigilance against biases and discrimination. Some fall into the category of high-risk systems under the AI Act, whose applicable rules in the employment field will come into effect on December 2, 2027.

Managing Interfaces Between Systems

The risk is not only in the tools taken separately. An AI, an information system, and security devices can each be correctly designed, yet create fragility during their exchanges. A poorly transmitted data point, overly extended access, or a rule interpreted differently from one system to another can be enough to create a vulnerability. The lesson from the Mars Climate Orbiter is also this: the error did not come from an isolated system, but from the interface between two systems that did not share the same convention.

Best Practices for AI Integration in HR

Mapping data, identifying the system that is authoritative, limiting access rights, and preparing a correction procedure are therefore deployment conditions and not just technical details. This approach aligns with the recommendations of the ANSSI, which advocates a risk-based approach for systems incorporating AI. Measuring AI by the improvement it brings These guarantees are not only about compliance or security: they also allow evaluating a project on its real value.

Evaluating AI's Real Impact on HR

The good criterion for an AI project in human resources is not the number of automated tasks, but its ability to improve, in good faith, a concrete moment of the lives of employees. This implies starting with an identified problem: a too complex procedure, a difficult-to-find information, an HR team saturated with repetitive requests. It is then necessary to check the quality of the data, set the limits of the system, and maintain a human validation when the consequences are significant.

Tracking Costs and Benefits of AI Projects

An AI project should also be evaluated on its real cost savings. For each use, the organization can associate costs (licenses, consumption, integration, human supervision, training, etc.) with measurable benefits (time saved, reduction of errors, improvement of response time, etc.). This return on investment is never fixed: the capabilities and costs of AI evolve rapidly. It must therefore be tracked use by use, then regularly re-evaluated to decide whether to continue, adjust, or stop a use case.

AI's Potential to Transform HR

Thus, AI can make HR services more efficient and more accessible, while helping organizations adapt more quickly to internal changes and market transformations. However, it is not enough to simply add AI to existing tools and practices to fully benefit from it. The challenge is to move from processing isolated transactions to accompanying the events that structure professional life: onboarding, mobility, salary evolution, skill development, or change in situation.

AI as a Support, Not a Substitute

AI can then help anticipate needs, guide paths, and coordinate actions at the right time. AI will not compensate for an imprecise HR policy, unreliable data, or the absence of human responsibility. Its value will depend on the ability of organizations to choose useful uses, preserve trust, and reinvest the time saved in transforming jobs and organization.