Publication date: January 16, 2026
Pharmaceutical production is one of the most regulated activities in Poland, Europe, and globally. The reason seems obvious: creating drugs can cause numerous complications, health problems, and in extreme cases, even death, depending on the individual patient’s health contraindications. Due to the above-mentioned reasons, the process of creating a new drug is often very expensive and lengthy, and involves a large number of professionals and scientists specializing in this field. In an era of dynamic digitalization, the need to develop specific operational rules for drug production and principles for using AI (Artificial Intelligence) and ML (Machine Learning) models is increasingly emphasized. Manufacturers are often considering implementing AI systems to support pharmaceutical production processes.
The European Medicines Agency has created a special report divided into drug production phases, highlighting the particularly useful aspects of AI-based models in each of these stages. The problem, however, lies in the ethics and risks of such use, due to the continuing ignorance and distrust of such algorithms, capable of predicting, learning, and processing enormous amounts of data in a short time. Due to these latter aspects, this tool—if used properly and ethically—can significantly improve workflow, but without appropriate legal regulations, it can also be very detrimental to potential patients. The purpose of this article is to present the pharmaceutical drug production process and cite basic legal regulations.with particular emphasis on the ethical use of AI systems in this process and the requirements in this respect imposed by EU and national law.
From the point of view of regulations that may be applied in Polish pharmaceutical production, it is necessary to mention primarily the Pharmaceutical Law[1], the provisions of the Civil Code on liability for damage caused by a dangerous product[2], GMP standards (Regulation of the Minister of Health of 9 November 2015 on the requirements of Good Manufacturing Practice[3]issued on the basis of the statutory delegation contained in the Pharmaceutical Law) and all recommendations related to the production of drugs (soft law). In the scope of the use of AI and ML models, it is worth mentioning Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence and amending Regulations (EC) No 300/2008, (EU) No 167/2013, (EU) No 168/2013, (EU) 2018/858, (EU) 2018/1139 and (EU) 2019/2144 and Directives 2014/90/EU, (EU) 2016/797 and (EU) 2020/1828 (Artificial Intelligence Act)[4], commonly referred to in the legal community as the “AI Act “, which, however, currently has the status of a pending act in some of its parts, which is justified by a certain expectation of the EU legislator for the smooth implementation of these regulations. to the methodology of everyday operations of enterprises (in the regulation, the “entity using” – Art. 3, item 4 of the “AI Act “). The regulation itself provides for very high sanctions, but also measures to support start- ups and SMEs (Micro, Small and Medium-sized Enterprises).
It seems reasonable to pay attention to the definition used by the AI Act to understand the characteristics of systems that actually define the scope of its standards for new technology systems. Article 3, point 1 of the Regulation provides a specific definition of an AI system, which, according to this definition, is a machine system designed to operate with varying degrees of autonomy after implementation, capable of adapting after implementation, and—for explicit or implicit purposes—inferring how to generate results based on received input data, such as predictions, content, recommendations, or decisions that may impact the physical or virtual environment. Therefore, models based on the operation of artificial intelligence and machine learning certainly fall under this definition, and their application falls under the standards of Regulation 2024/1689.
First and foremost, it’s important to highlight certain phases of drug development, adherence to which is a constant, necessary, and inherent element of the process. These stages are presented in the following, sequentially enumerated sections:
1. Drug discovery,
2. Nonclinical development,
3. Clinical trials,
4. Precision medicine,
5. Product information,
6. Manufacturing,
7. Post-authorization.
Before describing the specific phases of drug production, it’s important to outline the general requirements that legal regulations impose on potential manufacturers of medicinal products. As previously mentioned, the pharmaceutical industry is one of the most regulated sectors of the economy. A manufacturer must obtain a permit from the Chief Pharmaceutical Inspector to begin production. Only after obtaining this permit can they begin work on the new product and complete further stages of the procedure.
During the drug manufacturing process itself, it is crucial to adhere to Good Manufacturing Practices established in the Regulation of the Minister of Health of November 9, 2015, regarding the requirements of Good Manufacturing Practice. Furthermore, the use of AI and models based on machine learning should also be borne in mind, especially with regard to the regulation of so-called “prohibited practices”, to which these provisions are currently applied. These regulations have entered into force, and all companies using such computerized work are obligated to comply with the rigors contained therein. However, it seems best to discuss the potential application of these regulations, individualizing their fragments within a given phase of medicinal product production. Furthermore, manufacturers must also be mindful of product liability, which is regulated by the Civil Code and covers property and non-property damage caused to anyone by the product during its intended use. The basis of this liability regime is the principle of manufacturer’s risk, which means that the manufacturer’s fault in causing the damage is irrelevant.
Re. 1)
This process can be briefly described as a search for formulas and chemical compounds with therapeutic properties, as well as attempts to investigate the mechanism of action of a given disease, which allows for the selection of an appropriate biochemical combination for use in a drug to combat it. This involves creating an invention that, in the subsequent stages of non-clinical (point 2) and clinical (point 3) studies, will be tested for its response to the “target environment.” It is indicated that models based on artificial intelligence may find application in this regard in protein design and in the so-called docking process[5]. The European Medicines Agency (EMA) indicates the possibility of training AI and ML models to predict the most effective protein structures for use in treating a disease. Such positive application of new technologies in this field can significantly reduce the costs of developing a new medicinal product and its currently very long development time (current sources indicate that it takes several years for just one drug – this is due to the variability of reactions that can occur in a given patient’s body, which in turn necessitates multiple tests). However, the Agency also pointed out certain problems in its report, namely the possibility of so-called “black box” effect , which means that we have a certain output, but it is impossible to determine exactly how this result was achieved.
Good Manufacturing Practices require appropriately qualified individuals to participate in production from A to Z. There may be a risk of AI models examining the relevant characteristics of different scientists, as is the case, for example, in employee recruitment. Article 5, paragraph 1, point c (sub-items i and ii) of Regulation 2024/1689 prohibits the use of AI-based models to analyze personality traits, character traits, and even social behavior or emotions. In the case of a “recruiter”, such scoring may lead to discrimination or other unfair treatment of a candidate for work on the development of a new drug. Generally speaking, it seems that in this production phase, the greatest scope for illegal use of AI and ML may be the processes related to the examination of competences of human capital that could potentially participate in the production and supervision of this production, and therefore all models that could analyze data of a specific category (a concept under Regulation 2016/679 – GDPR[6]) and other non-personal data regarding candidates (e.g. propensity to commit crimes – also a practice prohibited in the field of AI under the EU act in question).
Ad. 2)
This stage primarily focuses on assessing the potential risks associated with using the drug in its current form, so that it can be modified appropriately if the results are unfavorable. In this stage, AI and ML are identified as tools that can enable greater humaneness in research by reducing the use of animals. Furthermore, AI can significantly reduce the cost of materials simply by reducing their use. After appropriate training, the AI, based on previous preclinical studies, will learn to predict the likely outcomes of new medicinal products during clinical trials and the risks that may arise with this use. The challenge, however, is to actually verify the probability of obtaining the correct result.
Ad. 3)
This phase of production involves testing and assessing the risk and effectiveness of a new medicinal product, albeit with human participation. This is where the broadest scope for AI and ML applications appears to be. It seems that new technologies and biomedical engineering products will play a significant role in the clinical trials phase, as it is necessary to meticulously analyze test results after administering a given drug sample to a patient. (With appropriate training and development of AI, it is conceivable that these models could, in principle, make probable diagnoses, thus narrowing the scope for further research for scientists working on the new medicinal product.) ML and AI models can also be used at this stage to select and collect data, generate average scores, and analyze them. Furthermore, the European Medicines Agency has indicated that AI can even aid in the study of emotional states in patients. All these areas of this innovation could ultimately help select the appropriate treatment for a specific group of patients while consuming significantly fewer resources—both time, human resources, and financial resources. However, the European Medicines Agency has also identified several areas where AI users face challenges. The protection of personal data and specific categories of data, such as biometric data, is once again at the forefront. Developers will need to ensure proper implementation of safeguards against data leaks (a requirement under the GDPR), and also be careful not to engage in prohibited AI practices, such as scoring or biometric data analysis, that are incompatible with its actual purpose (in this case, medical).
Ad. 4)
Of course, not every drug will work in the same way or with the same intensity in different organisms. In the fourth stage of medicinal product production, the need for individualization based on the patient’s genotype, allergens, chronic diseases, and other medical parameters that distinguish specific patients must be adequately assessed. In its report, the European Medicines Agency identified areas for the use of ML/AI, such as individualizing treatment conditions for each patient, adjusting dosage, and individualizing biomarkers[7] to be used during treatment. In this case, the risks associated with the use of AI may be most significant, due to the transition to clinical trials involving humans. Among the risks, the analyzed report notes the reluctance of potential patients to undergo therapies and treatments generated by AI.
Ad. 5)
The importance of this phase can be inferred essentially from its name itself. It involves creating specific documentation with its specific features, in order to materialize and consolidate the results of research on the product’s effectiveness and safety conducted in the preceding phases. AI can significantly accelerate the process of organizing information collected during the previous stages and generating such documentation. However, it should be remembered that the data contained in such documents is typically so-called sensitive data and cannot be processed by public products and systems based on artificial intelligence, as the processing process must be transparent. Furthermore, the documentation produced in this regard must meet the requirement of transparency (Chapter IV of the Regulation of the Minister of Health on the requirements of Good Manufacturing Practices sets numerous requirements for such information documentation). It also appears questionable whether it is possible to create documentation regarding the drug production process on an ongoing basis using AI at this stage, a requirement also included in the Regulation. However, under the AI Act, it seems reasonable to prohibit the use of AI models that create misleading content. The documentation – generated by AI – must certainly be properly checked by a team of specialists.
Ad. 6)
At the penultimate stage, a series of controlled biochemical reactions must be conducted (in accordance with the requirements of Good Manufacturing Practices listed in the Regulation of the Minister of Health referred to at the beginning of this text) to confirm the “practical” effectiveness, effects, and other parameters of the product. The Agency indicated that, according to Good Manufacturing Practices, the production of a medicinal product must be efficient, purposeful, and safe. Therefore, AI can be used to optimize the time and cost of the entire process, as well as to model and control the production process based on the analysis of current data. The agency also indicated that AI and ML can even monitor the maintenance of quality standards.
Ad. 7)
This is essentially a period, rather than a stage of medicinal production, during which the manufacturer obtains the legally required authorization to introduce the medicinal product to the general market and is available for use by a potential patient. AI is used in this area, particularly to analyze potential risks associated with drug administration, as well as to compress and analyze vast amounts of data, which would seem to require significant product control at this final stage. This can accelerate response times to risks identified during this phase and optimize countermeasures.
AI and ML appear to be tools that can help drug manufacturers combat diseases. However, it’s important to remember the legal requirements and enormous responsibility associated with this activity, particularly the provisions of the AI Act, the GDPR, and the Good Manufacturing Practices Requirements. Compliance with these requirements is a means of avoiding potentially significant sanctions from regulatory authorities, but also provides guidance on how to adhere to the “ethics” of medicinal product production.
[1]Journal of Laws 2025.750, i.e. of 2025.06.06;
[2]Civil Code (Journal of Laws 2025.1071, i.e. of 2025.08.06) – art. 449 1 -44910 ;
[3]Journal of Laws 2022.1273, i.e. of 2022.06.20;
[4]OJ EU.L.2024.1689 of 2024/07/12;
[5]a computer method that allows for the prediction of the preferred position of a ligand after binding to a macromolecule (e.g. a protein ) in its binding site to form a stable complex and for the interpretation of the interactions occurring between the bound ligand and the macromolecule.
[6]OJ EU.L.2016.119.1 of 2016.05.04
[7] Biomarker is a biological indicator , such as a substance, physiological property or gene , that indicates or may indicate the presence of a disease state or a physiological or mental disorder; it is also used to monitor the body’s response to therapeutic measures.