The rise of artificial intelligence (AI) and its widespread availability raise questions regarding its potential use in violating EU competition law. This issue is complex due to two characteristics of AI systems highlighted under the EU AI Act’s definition of such systems: (1) they operate with varying levels of autonomy and (2) they infer from the input they receive how to generate outputs such as predictions, recommendations or decisions that can influence physical or virtual environments.

This blog post explores the role of AI in anticompetitive horizontal and vertical agreements under Article 101 of the Treaty on the Functioning of the EU, as well as how AI could assist regulators in enforcing competition law rules.

Horizontal Agreements

AI may be used in cartels or hub-and-spoke arrangements. While autonomous price coordination does not violate EU competition law, it raises interesting questions for the future. Horizontal agreements in the AI market would be similar to those in other markets. However, due to AI’s high importance and growing demand, and the scarcity of talent and AI skills, no-poach agreements reducing competition in the labor market might be particularly noteworthy.

• Cartels. Explicit collusion is the clearest violation of EU competition rules, as competitors communicate directly to agree on anticompetitive practices such as price-fixing or market sharing. However, once an agreement is established, participants might deviate from the plan to achieve favorable outcomes for themselves. AI can be used to address this issue and facilitate the formation of stable cartels. Typically, cartel participants may use AI to automatically implement agreements, thereby reducing the need for direct communication. AI can also be used to monitor individual behaviors to ensure cartel stability. These cases do not present new legal challenges, as competition law rules apply as usual. The main difficulty lies in detecting the cartel and understanding the use of AI for such anticompetitive purposes. The issue is not new. In 2016, the UK Competition Markets Authority found in Posters that online sellers of posters and frames had used automated repricing software to monitor and adjust their prices and ensure that neither was undercutting the other. Cartel participants maintained contact to ensure the pricing arrangement was effective and to address issues with the operation of the repricing software.

• Hub and Spoke. Anticompetitive information exchange can occur indirectly, typically where competitors are aware that the price is set by a third-party AI-based platform and do not distance themselves from such practice. For example, in Eturas, travel agencies were suspected of applying a common cap on discounts through a third-party online booking platform. The Court of Justice of the EU confirmed that online platform terms setting a discount cap can lead to anticompetitive collusion with travel agencies. Travel agencies could therefore be presumed to have participated in such collusion if they were aware of anticompetitive amendments to the terms unless they distanced themselves. Indirect information exchange can also occur when competitors use the same third-party AI price-setting tool. The European Commission’s Horizontal Guidelines state that using a shared algorithmic pricing rule (e.g., matching a competitor’s price minus 5 percent) likely violates EU competition law, even in the absence of an explicit agreement. This view is untested under EU competition law, which requires an agreement. In Eturas, companies were found to have colluded because they knew of the anticompetitive practices using a common third-party tool and did not distance themselves. There was no violation where such knowledge could not be demonstrated.

• Autonomous Price Coordination. Competitors may independently employ distinct pricing AI tools using their own algorithm and datasets, through which they learn and adapt their price-setting strategies. Various experiments suggest that when such AI systems interact in a market environment, they tend to reach a price equilibrium that is higher than competitive prices. However, these experiments remain theoretical, and the evidence of algorithmic tacit collusion is limited. Competition authorities and academics are continuing to investigate this issue. Although tacit collusion does not currently fall within the scope of EU competition law, this issue might need to be reconsidered soon as AI becomes increasingly sophisticated and widespread. Rethinking tacit collusion could be theoretically possible, but it would be highly challenging, as it would involve questioning the core principles of EU competition law.

Vertical Agreements

Most vertical agreements, which involve competitors at different supply chain levels, do not breach EU competition law. The European Commission’s Guidelines on Vertical Restraints clarify that this is because the complementary nature of the activities performed by the parties involved in such agreements often means that pro-competitive actions by one party will benefit the other, ultimately benefiting consumers. However, certain vertical agreements may raise competition concerns under EU law.

• Input Foreclosure. There is potential for anticompetitive vertical arrangements resulting in the foreclosure of critical inputs to downstream players. Typically, if two firms in different segments of the AI supply chain agree to grant each other exclusive access to a valuable resource, it could hinder other competitors from developing competitive products. An example of this situation could be an AI chip manufacturer and an AI developer agreeing to provide each other with exclusive access to their respective semiconductor technology and advanced training datasets, foreclosing rival AI firms from obtaining these critical inputs. Under EU law, such an agreement would generally be presumed legal. However, if the market share of the parties to this agreement exceeds 30 percent, they do not benefit from this presumption.

• Hardcore Restrictions. Hardcore restrictions in vertical agreements are almost always illegal. Specifically, using AI to monitor or enforce resale price maintenance agreements, or exclusive or selective distribution systems, can violate EU competition law.

• Resale Price Maintenance Agreements. Sellers are prohibited from setting a fixed or minimum sale price for buyers. The increasing use of AI-driven price-monitoring systems by sellers in online markets enhances market transparency through price recommendations. However, these systems are not inherently illegal. Buyers still have the freedom to engage in competitive price strategies. The use of such systems only becomes illegal when buyers and sellers agree to turn recommended prices into mandatory ones. In such cases, the price-monitoring systems serve as enforcement tools for resale price maintenance agreements and are therefore illegal. For example, in 2018, the European Commission fined consumer electronics suppliers more than €110 million for using monitoring software to detect price deviations by retailers and intervene when prices fell, thus fixing retail prices. Many retailers, under pressure from electronics manufacturers to comply with the “suggested” prices, used algorithmic systems to automatically adjust prices based on other retailers’ prices, which affected overall prices more significantly than in offline contexts.

• Exclusive or Selective Distribution Systems. AI-powered monitoring mechanisms can serve as auxiliary enforcing tools for implementing exclusive or selective distribution systems. For example, AI can be used to monitor compliance with restrictions on the territory in which or the customers to whom the buyer or its customers may sell, or restrictions of cross-supplies between members of a selective distribution system.

Enforcement

Competition authorities are considering using AI to enhance case management and assist them in investigations by analyzing data and expanding e-discovery capabilities. This could help reduce the length of investigations, thereby limiting costs and uncertainty for companies under investigation. However, deploying AI for such purposes will likely take time, as it must be carefully designed and tested to ensure appropriate legal safeguards, including regarding the rights of defense, the right to good administration, and compliance with EU data protection and AI regulations, namely the GDPR and the AI Act.

The authors would like to thank Ioannis Dellis for his assistance in preparing this alert.