The webinar took a deliberately practical approach. Rather than starting with abstract legislation, it began with relatable HR cases and the question, “Is AI allowed to do this?”

The answer is often more nuanced than a simple “yes” or “no.” Under the AI Act, much depends on how you use AI, what impact that has on employees or candidates, and whether there is sufficient human oversight and transparency.

Whether you’re working with ChatGPT, Copilot, or a specialized HR tool, the level of risk depends on the specific application. Using an AI tool to draft a job posting more efficiently is therefore different from using that same technology to automatically screen candidates.

The AI Act uses a risk classification system:

• Prohibited applications:
  These are applications that run counter to European values, such as emotion recognition in the workplace or among job candidates. This prohibition is already in effect today.

• High-risk applications:
  These include, among other things, AI systems used in recruitment, as they can have a significant impact on a person’s career opportunities or livelihood.

• Limited risk:
  Think of chatbots. Here, the emphasis is primarily on transparency: users must know that they are interacting with AI.

• Minimal risk:
  For example, translation tools or applications with little impact on employees or candidates.

Kenny also distinguished between the provider of an AI tool and the “deployer” or user. In HR, the latter is often the employer: you may not build the tool yourself, but you do use it within your organization. And so you also share the responsibility for considering what that tool does, what data it uses, and what consequences it has for employees or candidates.

Imagine this: you receive 250 applications for a single job opening. To save time, you want to use an AI tool that analyzes all the resumes and automatically selects the most suitable candidates.

Is that allowed? Not without some caveats. AI can help you process information, recognize patterns, or pre-screen candidates. But the tool should not be the sole determinant of who moves forward in the selection process. After all, the impact on candidates’ career opportunities and access to employment is significant.

That’s why AI shouldn’t be allowed to make autonomous decisions here. There must be human oversight by someone with sufficient knowledge, capacity, and authority to critically evaluate the output and, if necessary, disregard it. In other words: AI can provide support, but HR remains responsible for the decision.

Imagine this: your organization has rolled out Copilot, or employees are using readily available tools like ChatGPT to write emails faster, summarize documents, or draft HR texts. It’s convenient, but everyone uses AI in their own way, without clear guidelines.

Is that allowed? In principle, anyone can work with AI, but not without conditions. Employers are still required to make efforts regarding AI literacy: employees must understand which AI systems are being used, what they’re for, how they work, and what risks are associated with them.

The recent Digital Omnibus Act has, however, redefined this obligation somewhat. Whereas the focus used to be on achieving an “adequate level” of AI literacy, the emphasis is now more on a duty of care. Employers must therefore be able to demonstrate that they are taking reasonable steps, for example by establishing an AI policy, providing training or awareness programs, and making clear agreements about what employees may and may not do with AI.

In practical terms, this means: you don’t have to train everyone to be an AI expert, but you must ensure that AI use does not occur in a completely uncontrolled manner. This is particularly important in HR, as employees often work with sensitive information.

Imagine this: your organization wants to make the selection process more objective. Instead of having different recruiters manually review all the resumes, you want to use an AI tool that screens all candidates in the same, consistent way. The idea: less gut feeling, less human bias, and a fairer selection process.

Is that allowed? Yes, AI can help reduce discrimination, but that doesn’t happen automatically. If the system is trained on historical data in which certain groups were underrepresented, AI can actually reinforce those patterns. For example, the webinar referenced the case of Amazon, where a recruitment tool was trained on 10 years of historical data that apparently contained a pattern favoring men’s chances of being hired; the tool adopted this pattern and consequently favored male candidates.

In practical terms, this means: don’t blindly trust the promise that AI is “more objective.” Ask vendors how their system was tested, what data was used, how bias is detected, and what documentation is available. AI can be a tool for fairer processes, but only if you actively verify that the output remains truly fair.

Imagine this: in a warehouse, forklifts are driving around and employees are working in close proximity to one another. To prevent workplace accidents, you want to have camera footage analyzed by AI. For example, the system could issue a warning when someone gets too close to a hazardous zone or when there is a risk of a collision.

Is that allowed? Yes, it is, but under strict conditions. The purpose is key here: using AI for safety is different from using AI to continuously monitor employees. As soon as camera footage is analyzed on the work floor, you must therefore very clearly define the system’s purpose, what footage or data is processed, how long it is retained, and who has access.

In addition, other rules continue to apply. These include national regulations regarding video surveillance, the introduction of new technology in the workplace, and other information and consultation requirements. And very importantly: in this context, AI must not be used for prohibited applications, such as emotion recognition in the workplace. So while safety can be a valid reason, “we want to analyze everything continuously because it’s convenient” is a completely different story.

Imagine this: your organization wants to implement a new AI tool to help HR better track workforce planning, performance, or absences. The tool seems efficient, the vendor promises clear insights, and HR is eager to get started right away.

Is that allowed? Not without further consideration. Even if an AI tool isn’t legally prohibited, existing rules regarding consultation and employee participation still apply. If the introduction of AI impacts terms of employment, pay conditions, work organization, or workplace well-being, you must determine which information and consultation obligations apply. The webinar referred, among other things, to the CPBW in this regard.

In practical terms, this means: do not treat AI as a purely technical implementation. Identify in a timely manner who needs to be involved, what questions or concerns employees and labor representatives may have, and how you will explain what the tool will and will not be used for. Even when no formal agreement is required—but information or advice is—it is wise to take that dialogue seriously. As an employer or HR professional, you can only learn from the concerns raised by a labor representative body. AI often touches on issues of trust, control, and transparency, and taking these into account is a key part of a proper and smooth implementation.

The rules surrounding AI continue to evolve, but that doesn’t mean organizations should wait. Several obligations are already in effect today, such as the ban on certain AI applications and the need to organize AI use in a conscious and responsible manner.

For HR professionals, it’s best to start with a simple but important exercise: identify which AI systems are already in use within your organization. Don’t just think of official HR tools, but also applications such as Copilot, ChatGPT, or AI features built into existing software.

Then, for each application, consider the following:

• What exactly are we using this AI for?
• What data is being processed?
• Does this impact employees or candidates?
• What risk category does the application fall into?
• Is there sufficient human oversight?
• Are employees, candidates, or social partners sufficiently informed?

Based on this, you can conduct a risk analysis and determine what measures are necessary. Consider clear agreements, training or awareness initiatives regarding AI literacy, documentation of human decisions, agreements with suppliers, and a policy regarding what data employees are or are not permitted to enter into AI tools.

Kenny also offered a practical tip: it’s best not to include your AI agreements directly in the employment regulations right away. Because the legislation is still evolving, a separate AI policy is often more flexible and easier to adapt.

The most important question regarding AI and whether it should be used to support HR is therefore not just “which tool should we use?”, but above all: “what will we use them for, what impact will they have, and under what conditions?”

Would you like more inspiration on how to apply AI in HR? Be sure to contact us.