Publication date: July 07, 2026
The dynamic development of artificial intelligence-based technologies is revolutionizing not only the commercial sector but also the area of state oversight of the digital market. The implementation of multi-agent systems by the Office of Competition and Consumer Protection (UOKiK) opens a new era in consumer rights enforcement, enabling the mass and automated identification of unfair market practices. With the Digital Services Act (DSA) and the Omnibus Directive in force, traditional control methods are giving way to algorithmic interface analysis aimed at eliminating so-called dark patterns and price manipulation. However, the use of “digital controllers” raises fundamental questions for legal science and business practice about the limits of automated decision-making processes in public administration. Although AI agents significantly improve the effectiveness of detecting violations, their legal status as a source of evidence remains the subject of heated debate. The main thesis is that while AI can be a powerful auxiliary tool for regulatory bodies, the ultimate responsibility for determining the facts and assessing the legitimate interests of a party must rest with humans, which is the foundation of a fair procedure in a state governed by the rule of law.
A key obligation of internet platform providers in light of modern regulations is to design interfaces in a transparent and ethical manner. The prohibition of manipulation, formulated, among others, in the Digital Services Act (Article 25), directly affects the structure of so-called deceptive interfaces (dark patterns). Websites and applications cannot be designed in a way that limits the recipient’s cognitive autonomy, interferes with their ability to rationally assess the situation, or forces them to make a purchasing decision that they would not have made under other circumstances.
One of the most glaring examples of such violations is the asymmetry in the contract conclusion and termination process, particularly evident in subscription models. This mechanism relies on extreme simplification of the purchase path while simultaneously mounting procedural barriers when attempting to cancel the service. Visual techniques are used here, among other things: payment activation buttons are highlighted with bright colors and a central location, while contract termination options are deliberately hidden at the bottom of the page, written in small font or masked with colors that blend with the background. Furthermore, canceling a subscription on online platforms often requires multiple selections or confirmation of the desire to cancel, despite the consumer’s prior explicit choice. Artificial intelligence algorithms, analyzing the page structure and visual hierarchy of elements, can pinpoint these disparities with mathematical precision, creating a list of violations that serves as hard evidence.
In the context of the Omnibus Directive, the obligation to disclose the lowest price 30 days before the discount has become a market standard, but its implementation is open to abuse. The practice of “empty promotions” involves artificially inflating the base price just before a planned discount or providing a false reference amount. In this area, AI agents demonstrate particular effectiveness, acting as real-time monitoring systems; they can archive the price history of each product, creating an independent database. Comparing this information with the entrepreneur’s declaration visible on the website allows for immediate detection of manipulation of the promotional algorithm.
An equally important area of control is the phenomenon of drip pricing , or hiding the real costs of a transaction until the final stage of the shopping cart. Businesses often employ a “decoy” strategy, presenting an attractive unit price, which, at the time of order finalization, is increased by mandatory, previously undisclosed costs, such as service fees, packaging costs, or payment processing fees. Pursuant to Article 12 of the Consumer Rights Act, businesses are obligated to clearly and understandably inform consumers about, among other things, the total price for the proposed service. Automated control systems are capable of conducting a full simulation of the purchasing process, from product selection to the payment gateway. Any discrepancy between the price presented in the product list and the amount required to complete the contract is reported by AI as an attempt to circumvent disclosure obligations and a direct violation of the collective interests of consumers.
According to Article 5 of the Act on Combating Unfair Market Practices, the key criterion for assessing a trader’s behavior is the impact of their actions on the recipient’s decision-making process. A market practice is considered misleading if “this action in any way causes or is likely to cause the average consumer to make a transactional decision that they would not otherwise have made”. The legislator specifies that both “spreading false information” and “spreading true information in a manner that is likely to be misleading” can constitute an infringement. In the digital environment, these manipulations most often focus on the “existence of a product, its type, or availability.” A common method of exerting unjustified pressure on consumers is the use of social proof mechanisms and an artificial sense of scarcity. This manifests itself in messages such as: “this product is now being viewed by x people,” “x items have already been purchased today,” or displaying timers indicating that “only 30 minutes left until the end of the promotion.” Particularly problematic from the perspective of trade ethics is the use of so-called false advertising. Timers – clocks counting down to the finale of a supposedly unique price opportunity. In reality, these are fake mechanisms, as after the specified deadline, the offer remains active and the product price remains unchanged or becomes even more favorable. This type of activity, a classic example of dark patterns, is designed to induce fear of missing out (FOMO) in customers and induce them to rush into a transaction. Using AI agents allows regulators to serially monitor such counters and prove their cyclical recurrence, providing direct evidence of deceptive practices.
With millions of transactions taking place across the country in just a few minutes or hours, standard order verification procedures prove insufficient to effectively fulfill the statutory responsibilities of supervisory authorities. Technological advancements in the form of AI algorithms come to the rescue. These algorithms can automatically monitor numerous commercial transactions simultaneously, generating preliminary opinions that are ultimately subject to human review. Such systems not only save significant processing time but, above all, enable oversight of a much broader range of businesses and their online platforms. The AI multi-agents used in this process are virtual “consumer robots” capable of mass-auditing e-commerce websites, simulating the natural behavior of online users to detect irregularities that a human controller would be unable to detect on such a large scale.
To conduct reliable and effective inspections, Polish law already offers supervisory authorities a toolkit in the form of the “mystery shopper” institution. Traditionally, this involves a person unrelated to the inspected company or the inspecting authority making a purchase and then completing a survey regarding specific activities they observe during standard shopping. The implementation of AI technology by the Office of Competition and Consumer Protection (UOKiK) aims to entrust AI multi-agents with the role of such digital “mystery shoppers.” Their task is to interact with the website interface, add a product to the cart, and complete the entire purchasing process without disclosing that this activity is being performed by an algorithm or that it is part of an official inspection procedure. This approach allows for direct verification of whether the entrepreneur is not using prohibited manipulative practices, known as dark patterns. However, it should be emphasized that the activity of AI multi-agents is strictly regulated by legal procedures and cannot be arbitrary. The algorithm operates under the strict supervision of the President of the Office of Competition and Consumer Protection, who, pursuant to Article 105ia of the Act on Competition and Consumer Protection, must always obtain prior consent from the Court of Competition and Consumer Protection. This mechanism serves as a key safeguard against abuse of power. Furthermore, after completing the inspection, the office is obligated to immediately provide the entrepreneur with an official ID and authorization for the inspection. In the age of digital administration, this obligation can be fulfilled electronically immediately after the AI multi-agents withdraw from the sales platform.
The key legal framework for the operation of algorithms commissioned by the regulator is provided by the EU AI Act. According to its provisions, AI systems used by public authorities for control and supervisory purposes should be considered high-risk AI systems. This entails a strict requirement to design them with appropriate transparency, which allows both the controlling and the controlled entities to properly interpret the system’s results and use them fairly. In practice, this means that algorithms must be built in an “explainable” model. A business subject to allegations based on an algorithmic audit has the statutory right to request full insight into the operation of AI tools. This transparency is essential for the controlled entity to understand the basis and criteria on which the authority deemed its online platform unfair or infringing on the collective interests of consumers (Article 24). This balance between the effectiveness of digital supervision and the right to defense is the foundation of a modern rule of law in the age of algorithms.
After completing the inspection activities on the entrepreneur’s online platform, the AI algorithm’s role evolves towards an analytical function, consisting of preparing an opinion indicating detected violations. In the context of potential proceedings against an entity employing unfair market practices, the admissibility of using such an analysis as valid evidence becomes a key issue. Pursuant to Article 7 of the Code of Administrative Procedure (hereinafter referred to as the Code of Administrative Procedure), which establishes the principle of objective truth, a public administration body is obligated to take all steps necessary to thoroughly clarify the factual circumstances. This obligation is consistent with Article 75 § 1 of the Code of Administrative Procedure, which introduces an open catalog of evidence, allowing as evidence anything that may contribute to the clarification of the case, provided it is not contrary to the law.
Under these regulations, the results of AI multi-agent work – taking the form of reports, opinions, or analyses generated after conducting an audit with court approval – fully fall within the statutory definition of evidence. However, it should be clearly stated that an AI opinion cannot be equated with an expert opinion within the meaning of Article 84 of the Code of Administrative Procedure. This stems from the fact that an algorithm does not possess the status of a natural person equipped with specialized knowledge, which is a statutory requirement for appointing an expert. Instead, documentation generated by an AI agent should be classified as a private document or so-called “unnamed evidence.”
Practical justification for this position can be found in the case law concerning digital evidence. The judgment of the Court of Appeal in Szczecin of September 19, 2016, I ACa 364/15, LEX no. 2147337 aptly describes this issue, pointing out that evidence in a case may include official and private documents, but also means other than those listed in Articles 305-308 of the Code of Civil Procedure. Electronic evidence, currently increasingly used in civil proceedings, is not explicitly listed in the catalog of means of evidence. However, the Code of Civil Procedure does not contain a closed list of evidence sources; anything relevant to the case may constitute evidence. Although the above ruling was issued in the context of civil procedure, due to the identical approach to the openness of the evidence system, it remains fully applicable to administrative proceedings conducted by the President of the Office of Competition and Consumer Protection.
The key element of algorithmic evidence remains the human factor, which serves as a primary safeguard over the autonomous operation of technology. It’s important to note that AI multi-agents, despite their high sophistication, operate based on statistical probability models, which carries the risk of misinterpreting dynamic website elements. For example, the system may incorrectly classify a standard technical error as intentional dark web activity. patterns or misinterpret the interface’s intentions in a specific cultural or linguistic context. Therefore, opinions generated by AI agents cannot constitute a standalone and final basis for a decision, but should be subjected to thorough, critical review by an official. Only such a comparison of the “raw” algorithmic result with human knowledge and experience allows for avoiding errors that could lead to unjustified penalties. This approach is directly supported by Article 80 of the Code of Administrative Procedure, according to which a public administration body assesses whether a given circumstance has been proven based on the entirety of the evidence. In this process, the “AI opinion” is only one of many components that must be weighed against other evidence and evaluated through the prism of principles of logic and life experience, ultimately guaranteeing the implementation of the principle of objective truth and protecting the entrepreneur from the automaticity of decisions made by the algorithm.
Multi-agent system implemented by the Office of Competition and Consumer Protection for automatic control of the e-commerce sector poses a significant challenge for entrepreneurs, forcing strict compliance with regulations regarding dark patterns, price transparency (Omnibus Directive, Art. 6a) and information obligations (Consumer Rights Act, Art. 12). These tools are used to mass detect manipulative practices such as drip pricing, fake timers or making it difficult to unsubscribe. Although AI agents perform a function similar to “mystery shoppers,” their activity must meet the rigors of Article 105ia of the Act on Competition and Consumer Protection, including the requirement to obtain court consent for a controlled purchase. What is crucial from a procedural perspective is that the findings made by the algorithm do not have the status of an expert opinion within the meaning of Article 84 of the Code of Administrative Procedure (lack of the status of a natural person with specialist knowledge), but constitute only a private document or “other evidence” subject to the authority’s free assessment (Article 80 of the Code of Administrative Procedure).
Consequently, the official is required to subject AI reports to thorough human review to eliminate the risk of misclassification resulting from so-called “AI hallucinations” or technical errors in the interpretation of the website’s code. The entrepreneur has full rights of defense based on the principle of active participation of the party (Article 10 of the Code of Administrative Procedure) and the principle of objective truth (Article 7 of the Code of Administrative Procedure), which means the right to question the bot’s logic and to access the instructions and parameters of the AI system, in accordance with the “explainability” requirement enshrined in the AI Act (Article 13). Any decision based solely on the automated generation of conclusions, without providing the party with an opportunity to comment on the evidence (Article 81 of the Code of Administrative Procedure), constitutes a gross violation of administrative procedure and may constitute an effective basis for challenging the authority’s decision.
Regulation 2022/2065 on the single market for digital services and amending Directive 2000/31/EC (Digital Services Act) (OJ EU L 277, 2022, No. 277, p. 1, as amended).
Directive (EU) 2019/2161 of the European Parliament and of the Council of 27 November 2019 amending Council Directive 93/13/EEC and Directives 98/6/EC, 2005/29/EC and 2011/83/EU of the European Parliament and of the Council as regards the better enforcement and modernisation of Union consumer protection rules (OJ L 328, 2019, p. 7, as amended).
Act of 30 May 2014 on consumer rights (consolidated text: Journal of Laws of 2024, item 1796, as amended).
Act of 23 August 2007 on counteracting unfair market practices (consolidated text: Journal of Laws of 2023, item 845).
Act of 16 February 2007 on competition and consumer protection (consolidated text: Journal of Laws of 2025, item 1714).
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) Text with EEA relevance (OJ L 1689, 2024).
Act of 14 June 1960, the Code of Administrative Procedure (consolidated text: Journal of Laws of 2025, item 1691).
Judgment of the Court of Appeal in Szczecin of 19 September 2016, I ACa 364/15, LEX no. 2147337.