AI Literacy as an HR Obligation: Training Requirements and Practical Programmes

Key takeaways
- Regulatory shift: Regulation (EU) 2026/1744 has replaced the original Article 4, moving the standard from guaranteeing specific individual knowledge to ensuring deployment efforts (modulos.ai).
- No mandatory certifications: The European Commission has clarified that there is no requirement for a uniform certificate or a specific level of AI literacy for every individual (osborneclarke.com) (modulos.ai).
- Enforcement status: National market surveillance authorities in EU Member States gained the power to audit and sanction non-compliant organisations on 2 August 2026 (protiviti.com) (modulos.ai).
- Risk-based approach: Compliance depends on the specific role of the employee, the associated risk of the AI system, and the individual's prior experience (modulos.ai).
As of October 2026, the regulatory landscape for Artificial Intelligence in the European Union has transitioned from theoretical debate to active supervision. Following the entry into force of Regulation (EU) 2026/1744 on 27 July 2026, the obligation for "AI literacy" has moved from a vague concept to a central pillar of organizational compliance (modulos.ai). National market surveillance authorities began actively supervising this obligation on 2 August 2026 (protiviti.com) (modulos.ai). For HR professionals, this means the question is no longer whether to train employees on AI, but how to structure that training to meet the specific requirements of the law.
The shift from result to effort
To understand the current HR obligation, we must look at the specific change in the legal text. The original Article 4 of the AI Act created a sense of urgency around ensuring employees possessed certain skills. However, the new Regulation (EU) 2026/1744 has fundamentally changed the nature of this duty (modulos.ai). The current text states that providers and deployers are not required to guarantee any specific level of AI literacy for any individual (modulos.ai).
This is a critical distinction for HR strategy. We have moved from an "obligation of result", where an organization might be held liable if an employee fails to understand a prompt or a risk, to an "obligation of effort." The law does not require you to measure the exact AI knowledge of every employee, nor does it require a "one size fits all" training programme (modulos.ai). Instead, the focus is on whether the organization has taken appropriate steps to provide the necessary skills, knowledge, and understanding to allow for informed deployment and awareness of risks (protiviti.com).
The mechanism here is risk-based. The German Bundesnetzagentur has noted that an organization's role, the associated risk of the AI system being used, and the staff's previous knowledge and experience are the decisive factors in determining what constitutes sufficient literacy (modulos.ai).
What the data says and what it does not say
When designing these programmes, HR leaders often look for benchmarks. However, we must be careful with how we interpret training data. The European Commission has stated that there is no need for a certificate to prove compliance with Article 4 (modulos.ai). This has significant implications for how HR departments report on "readiness."
What the data says: You can track participation rates in training modules, the completion of specific risk-awareness workshops, and the distribution of training hours across different departments (osborneclarke.com). These are measures of "effort" that demonstrate an organization is actively addressing its literacy obligations.
What it does not say: Data on training completion does not prove that an employee actually understands the risks of an AI system (modulos.ai). Also, because the law does not require measuring individual knowledge, a high completion rate in a Learning Management System (LMS) does not automatically insulate an organization from liability if an incident occurs (modulos.ai).
A lack of AI literacy can be viewed as a failure in the duty of care, particularly if that lack of understanding leads to an actual incident or harm (modulos.ai).
Why It Matters
Analysis: The liability gap
There is a tension between the "effort" standard and the reality of enforcement. While the law does not demand a specific knowledge score, the consequences of an error remain high. Analysis of the regulatory framework suggests that sanctions are more likely if there is evidence of an incident caused by a lack of appropriate training and guidance (modulos.ai).
While this shows "effort" and provides a paper trail of participation, it may fail if a high-risk AI system is deployed in a specialized department (such as recruitment or credit scoring) where employees lack the specific competence to oversee the system's output (modulos.ai).
Scenario B: The Risk-Weighted Approach An organization identifies that only 5% of its workforce interacts with "high-risk" AI systems or provides human oversight for automated decisions.
The practical implication is that HR cannot treat AI literacy as a single line item in the annual training budget. It must be segmented. For those tasked with human oversight under Article 26(2), the requirement is much higher: they must possess the necessary competence, training, and authority to oversee the system (modulos.ai).
The human oversight requirement
We must distinguish between general "awareness" and "competence for oversight." Article 26(2) requires deployers to assign human oversight to people who possess the necessary competence, training, and authority (modulos.ai). This is not a suggestion; it is a requirement for those managing AI systems. For certain high-risk systems (Annex III), these provisions become even more critical as we approach the 2 December 2027 deadline (modulos.ai).
The mechanism here is the "human-in-the-loop." If an AI system makes a decision that affects an individual's rights, and the person overseeing that system does not have the training to understand why the AI reached that decision, the organization is vulnerable. The "effort" is not just about teaching people how to use the tool, but teaching them how to challenge it.
| Feature | General AI Literacy (Art. 4) | Human Oversight Competence (Art. 26) |
|---|---|---|
| Primary Goal | Awareness of opportunities and risks (osborneclarke.com). | Ability to exercise authority and oversight (modulos.ai). |
| Target Audience | All employees and affected persons (osborneclarke.com). | Specific individuals assigned to oversight (modulos.ai). |
| Standard of Proof | Evidence of training "effort" (modulos.ai). | Evidence of "necessary competence" (modulos.ai). |
| Risk Level | Low to Moderate (modulos.ai). | High (specifically for Annex III systems) (modulos.ai). |
Practical programmes: A framework for HR
Since there is no "one size fits all" model, HR must build a framework that scales based on the user's proximity to the AI system. We should move away from the idea of a single "AI Certification" and toward a tiered competency model.
Tier 1: General Awareness (All Employees)
This is the baseline. The goal is to ensure every employee understands that AI is being used, what the general risks are (such as bias or hallucination), and how to report a concern (osborneclarke.com). This satisfies the "awareness" component of Article 3(56).
Tier 2: Functional Application (Power Users)
This is for employees who use AI as a daily tool (e.g., marketing, coding, or data analysis). Training should focus on "prompt engineering" as a skill, but more importantly, on verifying the accuracy of AI outputs to prevent the propagation of errors.
Tier 3: Oversight and Governance (Decision Makers)
This is the most critical tier for legal compliance. This is for the people assigned to human oversight under Article 26(2). They need training on the specific logic of the AI systems used by the company, the ability to intervene, and the authority to override an automated decision (modulos.ai).
There is no one size fits all for AI literacy and no strict requirements or mandatory trainings are imposed (modulos.ai).
Watch For
What HR can do on Monday morning
To move from a state of uncertainty to a state of compliance, HR teams should begin by mapping their internal AI landscape. We cannot train for risks we have not identified.
- Audit the deployment: Identify which departments are using AI systems and, crucially, which systems are classified as high-risk or fall under Annex III (modulos.ai).
- Map the roles: List the employees who are currently performing "human oversight" for any automated process. Check if their job descriptions include the "authority" required by Article 26(2) (modulos.ai).
- Define "Effort" for your firm: Create a training matrix that links training modules to risk levels.
- Establish a reporting mechanism: Ensure that the "awareness" part of literacy includes a clear, non-punitive way for employees to report when an AI system produces a suspicious or harmful result (osborneclarke.com).
The regulatory window for proactive adjustment is closing. While the EU has provided flexibility by not mandating specific certificates, the burden of proof remains on the organization to show that they have made a good-faith, risk-proportionate effort to educate their workforce. Compliance is not a certificate on a wall; it is a documented, tiered approach to competence that matches the risk of the technology being deployed.