Publication date: October 10, 2025
The development of artificial intelligence
Artificial intelligence is an interdisciplinary field of knowledge combining elements of computer science, mathematics, statistics, neuroscience, and cognitive science. Its goal is to create systems capable of performing tasks that previously required human intelligence. This includes the ability to learn from data, reason, make decisions, recognize patterns, and process and generate natural language. Unlike traditional programming, in which a computer executes strictly defined instructions, artificial intelligence aims to grant machines a degree of autonomy, allowing them to independently adapt their strategies to changing conditions. Today, AI is no longer an abstract theoretical concept, but a practical tool.
The development of artificial intelligence is one of the most dynamic phenomena in the history of science and technology. Its origins lie in simple deterministic algorithms, based on clearly defined logical and mathematical rules, used to automate repetitive calculations. However, the real breakthrough came in the second half of the 1990s, when, with the increasing availability of data and the development of computing power, machine learning began to be increasingly used. These methods allowed machines not only to perform pre-programmed tasks but, more importantly, to learn from input data and independently improve their results. The application of machine learning meant a shift from rule-based systems to statistical models capable of recognizing patterns and predicting future events. The next stage was the widespread adoption of deep learning, which gained practical significance in the second decade of the 21st century. These techniques utilize multi-layered neural networks capable of analyzing massive datasets in a manner similar to the perceptual processes occurring in the human brain. This enabled the recognition of images, speech, and natural language with unprecedented effectiveness. Deep learning is the basis of many modern solutions, such as recommendation systems.
The most recent phase of development is generative artificial intelligence, which became widely available around 2021. Unlike previous solutions, which focused on classifying and analyzing data, generative systems can create new content—text, images, sounds, and even complex economic strategies. The introduction of this type of technology has radically expanded the potential applications of AI, but it has also revealed new risks related to its impact on society and the economy. Generative models, capable of dynamically shaping information and influencing decision-making processes, can, for example, participate in market price manipulation.
Price manipulation with the help of new technologies
Price manipulation by sellers is unfair or illegal market practices that involve setting prices in a way that misleads consumers or restricts competition. One of the most commonly used strategies is price fixing. It involves agreements between independent businesses aimed at fixing or controlling prices. This can involve jointly setting minimum or maximum selling prices, coordinating price increases, or even fixing discounts. Such actions are automatically considered illegal – the mere fact of agreeing on prices is sufficient, even if the agreement is not fully implemented. The consequence of price fixing is the elimination of natural price competition. Consumers lose the opportunity to choose cheaper offers, and businesses lose the incentive to innovate.
Another form of manipulation is the abuse of a dominant position. A dominant position means that a business entity has an advantage in the relevant market, allowing it to operate largely independently of competitors, contractors, and consumers. Abuse occurs when a business entity uses its power to impose unfair pricing conditions. This can take various forms, such as setting prices that are excessively high relative to the value of the goods or using predatory pricing, i.e., underselling prices to eliminate competition. Each of these practices leads to market distortion, restricting access for new entrants, and worsening consumer conditions.
Pricing algorithms are computer programs that provide pricing recommendations or, in some cases, automatically adjust prices based on current and historical data on market conditions. These algorithms consider much of the same data that companies have always considered when making pricing decisions, including historical data as well as current supply and demand indicators. Compared to human price managers, algorithms can process significantly more data in a much shorter time. This efficiency allows companies using algorithmic pricing strategies to respond more quickly to changes in supply and demand and make pricing decisions based on a more accurate, real-time understanding of market conditions.
Ways to use algorithms
Algorithms can be used in various ways. One is to use the algorithm as a tool to achieve a goal set by the companies. In such a situation, the parties to the agreement make certain arrangements between themselves, and only their implementation is left to the algorithms. For example, two companies could agree to eliminate price competition between them. Simply using a software function that allows them to retrieve data on the prices of other market participants would be sufficient. A strategy would therefore be to automatically set prices slightly lower than other companies while simultaneously ignoring the prices of the colluding company. However, in this case, the mere fact of detecting the existence of such collusion does not raise any doubts under current competition law, as it is possible to attribute the concept of an agreement to this situation.
Another technique involves the joint use of a single price-response algorithm by several traders. This shared use of the same algorithm can lead to price alignment and less competition. One element of this type of agreement is the awareness of the participants that the information being shared will reach their competitors. Consequently, transparency regarding the future behavior of competitors is established between the recipients. Recipients are aware of the participation of their competitors at the same level of trade in the agreement and that their prices are disclosed to their competitors. However, in such cases, the use of algorithms remains a technical activity and can be assessed through the lens of the underlying conduct, to which existing provisions on anticompetitive agreements can be applied.
However, there does not always have to be any agreement between market players regarding the use of artificial intelligence in sales to raise legal concerns. An example would be the configuration of these modern tools to automatically respond to changes in competitors’ prices. This can take the form of faster price reductions, mimicking price increases, or simply adapting to current market price levels. The increasing use of specialized software by businesses that monitors websites and collects price data is making the online sales market increasingly transparent. This transparency allows for easy tracking of competitors’ pricing policies, quick detection of deviations from established price levels, and immediate response to such situations. This allows companies using algorithms to adjust their prices to rivals’ actions almost in real time. Consequently, traditional price reductions aimed at attracting customers often lose their purpose – competitors can offer the same reduction in a fraction of a second. The situation is different with price increases – if one seller decides to increase prices, others are likely to follow suit, leading to an overall price increase. As a result, prices naturally drift towards a level higher than fully competitive.
The most theoretical scenario is a situation in which self-learning algorithms, used independently by different companies, independently conclude that joint coordination is the most profitable strategy. This does not imply a formal agreement or explicit exchange of information, but rather the spontaneous development of behaviors akin to collusion. This could be facilitated by a combination of two factors: on the one hand, the vast amount of available data on competitors and consumers (e.g., thanks to the Internet of Things, transaction analysis, or online behavior tracking), and on the other, the growing capabilities of artificial intelligence algorithms that learn market strategies through experience.
Experimental environments have already demonstrated that algorithms using reinforcement learning are capable of developing stable pricing strategies that reduce competition, even when they are not explicitly programmed to do so. Research shows that such systems can gradually coordinate their behavior, balancing between exploration and exploitation of the environment in a manner resembling tacit collusion. Importantly, this phenomenon can occur even under conditions that hinder cooperation – for example, when new players enter the market or when demand fluctuates.
What is an AI algorithm?
An AI algorithm is a set of advanced mathematical rules and processes designed to solve tasks, make decisions, or imitate human behavior and thinking using a computer. An AI algorithm often leverages machine learning capabilities to analyze, process, and learn from data. This allows AI tools to more efficiently perform various tasks (predicting patterns, assessing trends, optimizing processes, etc.) that would otherwise require human intervention. AI algorithms form the foundation of artificial intelligence systems, enabling them to learn, reason, recognize patterns, process natural language, and make decisions.
AI algorithms can self-improve by adapting their actions based on the analysis of vast amounts of data. There are three major categories of AI algorithms: supervised learning, unsupervised learning, and reinforcement learning. The key differences between these algorithms lie in how they are trained and how they function.
Competition concerns – communication of algoritms
A problem that raises particular concerns among competition authorities is the potential ability of algorithms to communicate. Although there is currently no evidence that systems learn this type of interaction without human intervention, it theoretically cannot be ruled out that in the future, algorithms will develop their own mechanisms for exchanging information. The European Commission points to the risk of “novel forms of coordination” between computer systems. If such a situation were to occur, it would be easier to classify the companies’ actions as prohibited cooperation – similar to traditional information exchange between competitors.
The issue of legal liability, however, remains controversial. If algorithms merely anticipate competitors’ reactions and adapt their own strategies accordingly, companies can be said to have permitted market autonomy (pursuant to Article 101 of the TFEU). However, it is more difficult to assess cases in which the systems themselves create a “communication channel” leading to actual price coordination. Some legal doctrine proposes adopting an approach similar to that used in the case of unauthorized employee actions – the company would be liable for the tools used. This was aptly put by EU Competition Commissioner, emphasizing that “companies cannot hide behind computer code[1]“.
The current state of technology indicates that algorithms are not yet capable of concluding lasting cartel agreements in dynamic market conditions. However, the rapid development of artificial intelligence and the increasing complexity of predictive systems may enable the emergence of such practices in the future. Therefore, regulators are increasingly emphasizing the need to modernize competition law tools to effectively counter not only classic cartels but also “algorithmic collusion”. Failure to address this could lead to significant losses for consumers, a reduction in innovation, and the concentration of economic power in the hands of a few technological entities – creating a kind of “digital plutocracy.”
The effects of AI algorithmic pricing
One of the key problems with algorithmic pricing is the asymmetry in the speed of response to market changes. Companies with more advanced algorithms can update prices much more frequently—even continuously—while those with less advanced technological tools only make price adjustments at longer intervals, such as weekly. This leads to a structural competitive advantage for the former, as they can flexibly adapt to supply and demand and react almost immediately to competitive price movements. In practice, this means that companies with more advanced systems can aggressively lower prices before competitors have time to adapt their offerings, effectively driving them out of the market. This imbalance not only distorts the principles of fair competition but also leads to deeper market concentration, as smaller or technologically weaker companies gradually lose the ability to maintain their position against dominant players investing in advanced algorithmic solutions.
Predatory pricing, in which a dominant firm incurs short-term losses by deliberately pricing goods and services below cost to eliminate competitors or new entrants, is another practice that modern technology is currently employing. This strategy typically involves two stages: first, the dominant firm aggressively undercuts competitors’ prices to drive them out of the market (the predation phase), and then uses its market power to raise prices to recoup losses and generate profits after the competitors disappear (the loss recovery phase). For predatory pricing strategies to be effective, a firm must maintain low prices long enough to eliminate competitors. Pricing algorithms can help firms target specific customers of competitors by offering them prices even below cost. For example, an established firm might do this to avoid losing customers to a new competitor. An established firm might use an algorithm to target customers most likely to switch suppliers, seeking to retain them rather than offering lower prices to all its customers. This could help an established company minimize losses. These algorithms can also help companies pursue predatory pricing strategies and build a reputation for lowering prices in the future if new entrants struggle. Pricing algorithms can also help companies raise prices for consumers who are more willing to pay or less sensitive to price changes. They enable companies to simultaneously engage in predatory pricing and recover losses without the need for a human intervention, using automated means.
Finally, algorithmic pricing introduces a significant element of uncertainty for consumers, who are unable to predict how much they will ultimately pay for a product or service. These mechanisms, based on automated supply and demand analyses, lead to constant and opaque price fluctuations, undermining market trust and limiting the ability to make rational purchasing decisions. Price instability often causes consumers to feel compelled to buy quickly for fear of further cost increases, which encourages impulsive and economically unfavorable choices. In the long term, such practices destabilize the market, hinder healthy competition, and strengthen the position of dominant players who exploit technological advantages at the expense of weaker market participants.
The question of legality
However, a fundamental question arises regarding the legality of such practices under Polish law. The Act of 9 May 2014 on Information on the Prices of Goods and Services[2] imposes on businesses the obligation to clearly and unambiguously disclose the prices of their products. According to the Act, the price should be clearly displayed and allow for comparison with other market offers. However, dynamic price changes, even occurring several times a day, can raise interpretational questions regarding compliance with the requirement of unambiguous price presentation. A consumer who sees significantly different prices for the same product or service within a short period of time may be deprived of the ability to make a rational economic choice. It should be emphasized that the legislator also introduced the obligation to disclose the lowest price applicable within thirty days prior to the discount. In the context of algorithmic pricing, the question arises as to whether every short-term price change resulting from the operation of an algorithm should be treated as a discount within the meaning of the Act, or whether it should be classified as a normal market fluctuation. The lack of clear regulations in this area leads to legal uncertainty for both entrepreneurs and consumers.
The Act of 16 February 2007 on Competition and Consumer Protection[3] opens up an even wider field of interpretation. This Act provides both instruments for counteracting practices that violate the collective interests of consumers and protection against anti-competitive practices. In this context, the problem of so-called tacit algorithmic collusion is particularly significant. In this context, independently operating algorithmic systems of various businesses, monitoring and reacting to competitors’ prices, can stabilize them at an inflated level without the need for a formal agreement. This type of phenomenon, although difficult to detect and prove, can be classified as a violation of fair competition principles, potentially subject to intervention by market protection authorities. The literature indicates that the increasing automation of decision-making processes in price setting raises the risk of developing anti-competitive coordination mechanisms that fall outside the traditional categories of antitrust law.
At the same time, the Competition and Consumer Protection Act imposes an obligation on businesses to avoid misleading practices. In the case of dynamic pricing, the lack of full transparency becomes a problem. If consumers are not informed upfront that prices may fluctuate significantly depending on demand or transaction time, they may be deemed misleading, and the practice itself may be deemed to violate collective consumer interests. In this sense, the obligation of transparency takes on particular significance and should also include information about the pricing mechanism, not just the current price level.
An analysis of the applicable regulations leads to the conclusion that algorithmic pricing per se is not a prohibited practice in Poland. However, its legality depends on whether the entrepreneur complies with the obligations arising from specific laws, as well as on whether the use of algorithms does not lead to practices that restrict competition or violate consumer rights. Entrepreneurs are obligated to provide consumers with reliable and complete information and to avoid practices that may destabilize the market. In the event of a violation of these obligations, sanctions may be imposed by both the Trade Inspection Authority (in the case of incorrect pricing information) and the President of the Office of Competition and Consumer Protection (in the case of practices that violate competition rules or are misleading).
Case law review
It is also worth taking a closer look at decisions issued on algorithmic pricing by state and EU authorities.
In the Eturas case, the contested agreement was supported by a digital platform (software for selling travel online), where the system’s administrator proposed to competing travel agencies the use of a technical instrument imposing a ceiling on discounts on packages offered. The EU Court found it reasonable to assume that travel agencies that were aware of the content of messages sent via the system were participants in an anticompetitive agreement, unless they rebutted that presumption. The Commission also states that the prohibition under Article 101(1) TFEU is likely to cover cases in which “pricing rules” were defined by undertakings “in a common algorithmic tool (e.g., rules for adjusting the price to the lowest price on a specific online platform or in a specific online store), and this qualification would be accepted “even in the absence of an express agreement to adjust future prices”[4].
In commercial transactions, there are also cases in which entrepreneurs use the same algorithm to set prices for their services, yet there are no grounds to conclude that they have entered into an anticompetitive agreement. An example of a case assessed by the President of the Office of Competition and Consumer Protection (UOKiK) is the UBER app. Its use does result in the restriction or even elimination of price competition between UBER drivers, which is, after all, centrally determined by the app. This also constitutes an agreement between UBER and individual drivers. In this case, however, such algorithmic pricing is necessary for the proper functioning of the UBER system, and at the same time, it does not lead to the elimination of competition, being a proportionate measure. This justifies treating such an agreement as not violating the prohibition of Article 6, Section 1 of the UOKiK based on the construction of ancillary restrictions.
A problematic scenario can arise when a large number of businesses utilize the same or similarly functioning algorithm without any collusion between them regarding the selection of one algorithm over another. In this situation, each of these businesses uses a specific program as a tool to inform them about the market situation and enable a rapid response. In this scenario, there is a significant increase in market transparency, which can have anti-competitive effects, as the use of an algorithm allows businesses to react much more quickly to any market changes, including price reductions by competitors, which can reduce their incentive to make such reductions. In such a case, businesses merely adjust their prices to those of their competitors. Such conduct – so-called parallel conduct – has for years been viewed by EU and Polish jurisprudence as not violating antitrust rules. In fact, in this case, it cannot be said that an agreement was concluded. Moreover, under the soft law provisions issued regarding vertical relationships, the Commission itself has indicated that price monitoring using computer programs is not prohibited, or more precisely, it may benefit from an exemption from the prohibition of anticompetitive agreements. Therefore, from the Commission’s position, it can be deduced that the antitrust permissibility of such algorithmic price monitoring is determined by the lack of grounds for attributing such conduct the status of an agreement, and that it constitutes a manifestation of the parallel conduct mentioned above.
According to the authorities, it would also be possible to attribute antitrust liability to the creator of the algorithm. The VM- Remonts formula, developed in EU case law, could be particularly applicable to this. In this case, while the Court recognized as a rule that an undertaking cannot “be held liable for participating in a concerted practice on the basis of the actions of an independent service provider”[5], it also identified exceptions to this rule. Fulfilling one of these exceptions justifies the application of Article 101(1) TFEU also to the service provider, for example, the software developer. Polish legal literature also argues that the constructions of extended liability for competition law infringements developed in EU case law (especially the concept of cartel accessory liability) could also be used to attribute antitrust liability to an undertaking that provides participants in an agreement with tools (including digital tools) enabling the conclusion or implementation of a prohibited agreement.
Summary
Dynamic, algorithmic pricing lies at the intersection of two key areas of law: consumer law and competition law. This practice, while permissible, requires particular caution on the part of businesses and the development of consistent interpretative guidelines by the legislature. The lack of clear regulations creates a risk not only for consumers, who may be exposed to non-transparent and unfair practices, but also for businesses themselves, who may suffer severe financial consequences if they violate fair competition regulations. Given the growing importance of algorithmic technologies, it seems necessary to further develop and clarify the legal framework to enable businesses to use innovative price management tools, while also ensuring an adequate level of consumer protection and the integrity of market mechanisms.
The most serious challenge for competition authorities is when algorithms maintain elevated prices without formal information exchange between companies. In such cases, the classic dogmatics of antitrust law may prove insufficient, and consumers will bear the cost in the form of a loss of some of their well-being. Therefore, developing technical competencies of supervisory authorities, enabling them to understand and analyze the mechanisms of algorithms used in business practice, is particularly important.
At this stage, it seems premature to introduce new regulations on algorithmic pricing, as this could hinder the development of innovative technologies and limit their beneficial applications. Further interdisciplinary research combining law, economics, and computer science, as well as careful observation of market practice and case law, is essential.
[1] Speech of the European Commissioner for Competition M. Vestager, Berlin, 16/03/2017.
[2] Act of 9 May 2014 on providing information on prices of goods and services (consolidated text: Journal of Laws of 2023, item 168).
[3] Act of 16 February 2007 on competition and consumer protection (consolidated text: Journal of Laws of 2024, item 1616, as amended).
[4] Point 397 of Guidelines 2023/C 259/01.
[5] Point 33 of the judgment of the Court of Justice of July 21, 2016, C-542/14, SIA “VM Remonts” (formerly SIA “DIV un KO”) and others. v. Konkurences padom, ECLI:EU:C:2016:578.