AI in HR: enormous potential, growing responsibility

Imagine a situation in which AI screens hundreds of job applications in a matter of minutes and recommends the best candidates for interview, or optimises a complex shift schedule while taking into account the needs and preferences of thousands of employees. This is no longer distant wishful thinking, but a reality in many organisations. According to research, 52 per cent of Finnish organisations use AI solutions in HR tasks. Although adoption is still fragmented, the potential is enormous: AI can make recruitment, onboarding and offboarding processes more efficient, facilitate the allocation of shifts and tasks, support performance management and produce more advanced people analytics.

However, AI is only as capable as its users. Using intelligent tools effectively requires HR professionals and managers to be trained and to understand the applicable ground rules so that the benefits can be achieved responsibly and sustainably. At the same time, more is required from employers: the use of AI raises legal, ethical and practical questions that cannot be resolved merely by producing an AI strategy document. Legislation is imposing increasing obligations on the use of AI, most recently through the EU AI Act, which sets requirements for both providers and users of AI applications.

Introducing AI: needs, risks and employee participation

An employer must address several important issues before introducing an AI system into HR management. The first step is to identify the need for which AI is intended to be used and assess the associated risks. Employment and data protection legislation in particular require employers to assess in advance how new technology may affect employees. If, for example, an AI-based tool supporting recruitment is to be introduced, this concerns the very core of processing job applicants’ personal data. The employer must identify and justify why personal data is processed, how and to what extent it is used, and what changes the AI system may require in existing practices or in the information provided to applicants. Data protection legislation requires, among other things, an assessment of risks relating to personal data, including a data protection impact assessment where applicable, before new technology is introduced, regardless of whether the technology uses AI. In practice, the use of AI systems in HR almost automatically means that risks to employees’ privacy must be identified and the necessary safeguards determined in advance. Fully automated decision-making in recruitment, such as screening job applications without any human involvement in the final decision, is generally prohibited under data protection legislation. The recruiter must therefore always retain a meaningful role in the process.

Employers also have cooperation obligations when the organisation’s operations are developed by introducing new technology. Every organisation employing at least 20 people must engage in dialogue with its personnel to safeguard employees’ opportunities to influence matters affecting them. The introduction of AI-based solutions in HR is clearly such a matter and should be discussed with personnel well in advance. Employers with more than 50 employees are also subject to an express, simplified change negotiation obligation when introducing new technology. If AI is expected to reduce the need for labour or materially alter employees’ duties, more extensive change negotiations must be conducted with personnel before implementation. All these cooperation procedures must take place at the appropriate time, meaning before procurement decisions are made, in order to meet the requirements of the Act on Co-operation within Undertakings.

The Occupational Safety and Health Act also applies to the introduction of AI in the workplace. A central principle of the Act is that employers must identify work-related hazards and harmful factors and address them proactively. AI may introduce new dimensions to traditional occupational safety considerations: what kinds of risks and strain may arise from its introduction, and how should they be prevented? Learning to use new technology may, for example, place a psychological burden on employees, while concerns about their rights may cause stress when AI becomes involved in HR processes and managerial work. Employers must assess these risks as well and provide appropriate induction, support and measures to safeguard employee wellbeing during the change.

It is already widely recognised that the use of AI presents challenges for non-discrimination in working life. Under the Non-Discrimination Act, an employer may not treat employees or job applicants differently on discriminatory grounds such as age, gender or another personal characteristic. Because AI learns and draws conclusions from the data provided to it, it may absorb biases hidden in that data. A recruitment algorithm may, for instance, favour applicants on the basis of gender rather than merit. If most people recruited in the past have been of a particular gender, the AI system may interpret this as a “model of success”. Eliminating discriminatory bias from AI is difficult because algorithms are often opaque to users and their decision-making may be difficult to explain. On the basis of the current Government Programme, a research project has been launched in Finland’s public administration to identify and prevent discrimination risks associated with AI.

The EU AI Act will gradually tighten the requirements

The European Union has also entered the field of AI regulation. The EU AI Act entered into force in summer 2024 and introduces new requirements for the use of AI. Since February 2025, organisations have already been required to ensure that their personnel have an adequate level of AI literacy. AI literacy refers to employees’ ability to assess AI-generated outputs critically and to use AI responsibly and appropriately.

Further obligations will follow from August 2026, when the core risk-based requirements of the AI Act begin to apply. The Act distinguishes between four categories: prohibited AI practices, high-risk systems, limited-risk systems and minimal-risk systems. This will be a significant milestone for employers using AI, as many systems acquired to support HR processes are classified as high-risk AI systems under the Act. A high-risk classification also brings more extensive statutory obligations for the employer using the system. Compliance with the AI Act is reinforced by substantial administrative fines, which may amount to millions of euros depending on the size of the company.

Applications used to recognise employees’ emotions are prohibited where they analyse matters such as an employee’s intentions or job satisfaction. The use of biometric identifiers to categorise individuals on the basis of ethnic origin, political opinion, religion or sexual orientation is likewise prohibited. Social scoring based on personal characteristics is also prohibited where it results in detrimental treatment, such as restricting or preventing career progression. Subliminal manipulation and the exploitation of vulnerabilities are not permitted either.

In the terminology of the AI Act, an employer will typically be the deployer of a system where it acquires a ready-made AI solution for HR purposes. However, an organisation may modify a general-purpose AI system for its own purposes, in which case the employer may become a provider under the Act. The distinction between the roles of deployer and provider is important: providers are subject to significantly broader legal obligations, including continuous quality assurance, technical documentation, system certification and detailed regulatory reporting, compared with a deployer. Employers should therefore generally prefer ready-made applications designed for HR use and use them strictly in accordance with the provider’s instructions and intended purpose.

Higher risks and greater responsibilities

What types of AI use are considered high-risk in an HR context? Automated decision-making and profiling based on personal characteristics are always high-risk. Under the AI Act, high-risk systems include those that affect access to employment, employment terms, career progression or decisions concerning performance at work. Examples include AI systems used in recruitment, decisions concerning employment conditions and career development, or the termination of employment. The same category includes systems that allocate work tasks on the basis of a person’s behaviour, personality or other personal characteristics, as well as systems used to monitor and assess employee performance during employment. These are all situations in which AI directly affects individuals and their treatment in working life. They are precisely the kinds of sensitive situations in which risks must be identified.

The dividing line is not always entirely clear. Some of the uses described above may be considered low-risk if they do not cause significant harm or pose risks to employees’ health, safety or fundamental rights and do not materially influence decisions concerning them. For example, a system that screens job applications and recommends the most suitable candidates to a recruiter would clearly appear to be a high-risk application. By contrast, an AI tool that merely classifies and transfers applications between systems without influencing whether applicants progress in the recruitment process, or that identifies anomalies in decision-making without intervening in the decision itself, would fall into the minimal-risk category. Interactive AI tools, such as virtual assistants used in HR matters, fall into the limited-risk category. In such cases, the employer must inform users that they are interacting with AI.

When an employer introduces a high-risk AI system, the AI Act requires it to fulfil several obligations. First, it must ensure that the system is used appropriately and in accordance with its instructions and intended purpose. Second, the organisation must appoint a responsible person or team to oversee the system’s operation. Those responsible for oversight must have sufficient competence, training, authority and resources to perform the task. The employer must also pay particular attention to data governance and ensure that the data entered by the organisation is relevant to the intended purpose and sufficiently representative so that the system does not produce misleading results. In addition, the user organisation must respond to risks arising during use and report any errors or biased decision-making to the system provider and the competent authority where required, and cooperate with supervisory authorities where necessary.

Transparency is another key element of the Act: employees must be informed when an AI system is introduced, and where they are subject to AI-based decisions, they have the right to receive an explanation of a decision affecting them. Last but not least, the employer must retain the AI system’s automatically generated logs for at least six months where those logs are under its control. The logs may contain information on how the system reached particular outputs or decisions. Retaining and tracing this information is essential when investigating possible disputes at a later stage.

As noted above, existing legislation already requires employers to follow similar principles in many respects. Employees may, for example, be informed appropriately through cooperation procedures, while data protection legislation already requires personal data to be processed appropriately. The EU AI Act nevertheless adds new and concrete AI-specific obligations: when using high-risk AI systems, organisations must, among other things, retain logs and ensure continuous and adequate human oversight throughout the system’s lifecycle.

Users and data at the heart of an AI strategy

Although legislation provides the framework for responsible AI use, its ultimate success depends on people. A company may create an ambitious AI strategy, but if employees lack the ability, willingness or confidence to use the system, the expected benefits may not be realised. User trust is crucial, and in decisions affecting personnel it is critical. Research indicates that concerns about employees’ rights or the purpose for which a system is used directly affect whether employees accept AI as part of their everyday work. Trust is similarly weakened by concerns that employees’ ability to influence matters may be reduced and by doubts regarding the system’s actual capabilities.

Employers should therefore invest in introducing AI as openly and transparently as possible and in providing comprehensive information to personnel. Integrating AI into HR functions is always a process of change that requires traditional change management and learning. New technology may initially place a burden on both its users and those affected by its decisions, depending on the individual’s role and readiness. Management should therefore listen to employees, provide the necessary support and, above all, communicate clearly about the change. Subsequent disagreements and potential legal proceedings are also best prevented by investing in competence and AI literacy.

Usability is central to a successful AI investment. A phenomenon known as “shadow AI” has emerged: where employer-provided tools are perceived as difficult or inefficient, employees may begin to use external AI solutions instead of the organisation’s approved tools. This may cause the benefits of the investment to be lost while creating concrete data protection and cybersecurity risks as data is processed through uncontrolled channels. In addition to ensuring adequate AI literacy, organisations must therefore provide proper user training so that everyone knows how to use the new systems. Clear internal guidance is also needed on how external AI tools, such as general-purpose chatbots and other services, may be used at work. This helps protect personal data, trade secrets and other confidential information.

A central principle of the AI era is that an AI system is only as reliable as the data it uses. In addition to considering big data, organisations must pay close attention to their own HR data. If the data is incomplete or inaccurate, the AI system will inevitably draw incorrect or imprecise conclusions. At worst, an employer may unintentionally discriminate against an employee or job applicant where historical data contains structural bias. The EU AI Act therefore expressly requires the quality and reliability of data to be ensured in high-risk HR systems. In practice, this means that HR data must be collected, updated and cleaned continuously so that AI-based decisions are based on information that is as representative, accurate and relevant as possible. This is not an entirely new principle, as Finnish employers have for more than twenty years been required under the Act on the Protection of Privacy in Working Life to process only employee data that is necessary for the employment relationship.

Conclusion

The enormous potential of AI will undoubtedly be used more extensively to support HR functions in the coming years. Although its use is not free from legal or ethical challenges, AI can at its best make work more meaningful and HR and managerial processes more equal, personalised and efficient. However, AI must be introduced and used systematically so that organisations can realise its full potential while safeguarding employees’ rights. Ultimately, people remain at the heart of all of this: the users who turn strategy into reality in their daily work. Organisations that genuinely invest in their personnel’s technical and ethical competence, as well as in user-friendly tools, will be well positioned in the age of AI.

Kaisa Salo

Counsel

kaisa.salo@legalfolks.fi

+35840 168 1418

To receive our articles directly by email, subscribe to the Folks newsletter here.