From Human Cartels to Digital Coordination: Rethinking Section 3(3) in Algorithmic Markets
Introduction
Traditionally, competition law has viewed cartels as the product of deliberately coordinated human activities. These include activities such as price-fixing, market allocation, limiting output, and bid-rigging. Cartels have historically required the existence of explicitly articulated agreements or tacitly agreed upon actions through communication and conscious coordinated strategic actions among competing businesses. Therefore, within this interpretation of cartels, Section 3(3) of the Competition Act, 2002 prohibits agreements made by businesses that are in the same or similar lines of business, presumes that any such agreement would result in an appreciable adverse effect on competition (AAEC), and thus punishes such conduct. In both India and globally, the enforcement of cartel conduct has focused on identifying intent, demonstrating concerted action, and establishing communication through either direct (overt) or circumstantial (covert) evidence such as parallel conduct supported by plus factors.
However, in industries where pricing algorithms (or AI-based pricing systems) are used to set prices (like e-commerce, ride-sharing services, airlines, hotels, and e-commerce), the anthropocentric framework is increasingly tested. In sectors such as aforementioned, firms use pricing algorithms that monitor current market conditions continuously (either directly or indirectly), observe the pricing of its competitors, and set price levels accordingly, within real-time intervals. Machine-learning algorithms in particular are capable of generating sustained, parallel pricing results in the absence of direct (overt) communication and conscious coordination. As such, some markets will show cartel effects (e.g., higher average prices, less dispersion in pricing among competitors, lower levels of competition) even when conventional evidence of an agreement does not exist.
The advancement of algorithms poses a key challenge regarding the interpretation of the Competition Act as it relates to Section 3(3). When determining whether the coordination through algorithms could fit under section 3(3) of the Competition Act 2002. The relevance of this issue is important as the Peak AAEC assumption is only available where there is an agreement, as notified in section 2(b). The uncertainty surrounding E-commerce markets where coordinated behaviour could happen unintentionally, as well as the lack of existing mechanisms for disciplinary enforcement, calls into question the current interpretation of the Act. Indian courts have consistently ruled that price parallelism alone does not contravene section 3 of the Competition Act. In Rajasthan Cylinders & Containers Ltd. v. Union of India, the Supreme Court cautioned against equating conscious parallelism with cartelisation, recognising that similar conduct may naturally arise in oligopolistic markets. At the same time, decisions such asFx Enterprise Solutions India Pvt Ltd. v. Hyundai Motor India Ltd.1 show the Competition Commission of India’s readiness to infer agreements from indirect facilitation even in the absence of direct horizontal communication.
Table of Contents
To Reconceptualise Algorithmic Collusion: Human Interaction with Smart Machines
For many years, the basis for the study of Cartels (Collusion) is that the participants have intentionally worked with one another through some form of communication and mutual understanding prior to engaging in any conduct. Algorithmic collusion removes this assumption, providing that the pricing decision is made with minimal human intervention by an Automated System which is capable of Learning, Predicting, and Reacting to Market Behaviour. In order to determine whether Algorithmic Collusion fits within Section 3(3) we need to develop a framework that allows for distinguishing different types of Algorithmic Collusion from one another.
According to various entities including OECD, there are different models of human versus algorithmic involvement with respect to collusion. The first model is the messenger/executor model, where firms agree to collude on their own through mutual consent and then use algorithms to implement and monitor that cartel. This type of collaboration obviously meets the criteria outlined in Section 3(3) because it is based on mutual agreement between humans.
An alternative model for algorithmic collusion is the hub-and-spoke model, whereby a common platform or provider for algorithms coordinates competition between firms. While these firms do not speak to each other directly, the algorithms centralised nature fosters coordinated outcomes. In support of this model India’s jurisprudence view regarding indirect facilitation can be confirmed by the case of Fx Enterprise Solutions v Hyundai Motor India Limited where a vertical arrangement had been found to facilitate horizontal price coordination.
The most difficult model of collusion would be the autonomous algorithm model, where independently operating bots observing both markets and competitors learn that co-operation leads to a higher profit than fighting with each other. Such co-operation is produced through cartel-like outcomes without any prior agreement or communication and will be the basis for future litigation and testing of Section 3.
Algorithmic Collusion and Tacit Coordination
Autonomous algorithmic collusion may appear similar to lawful tacit collusion or conscious parallelism, which competition law tolerates due to the absence of agreement. However, this comparison overlooks key differences. Tacit coordination in oligopolistic markets is constrained by human limitations, strategic uncertainty, and fragile expectations. Algorithmic systems, by contrast, reduce uncertainty, react instantly to deviations, and stabilise coordinated outcomes over time. Artificial intelligence therefore increases the likelihood, durability, and effectiveness of coordination, resulting in greater consumer harm.
Under Indian competition law, the delineation of “independent actions” from “communications” helps to assess whether there’s a violation under section 3 of the Act. In addition, the mention of “parallel pricing” in the section implies great difficulty in proving illegal agreements for businesses that utilize algorithms when establishing their pricing strategies. If any conclusion is drawn regarding algorithmic parallelism, it will essentially undermine the possibility of successful licensing or prosecuting any unlawful agreement established between online entities.
Under Section 3(3) of the Competition Act, 2002
Proof of an “Agreement” under Section 3(3) does not require communication or intent, but instead focuses on whether an ‘Agreement’ exists (defined as an ‘agreement’ in Section 2(b) as any arrangement, understanding or agreement by two or more companies operating together). This broad definition allows for the possibility of prosecuting algorithmic collusion when companies knowingly and intentionally develop and utilize pricing algorithms that coordinate prices. The challenge now is to establish whether an independent algorithmic pricing arrangement has progressed from “unilateral conduct” to an “attributable agreement” for design, development and deployment of algorithms that produce coordinated prices amongst competing companies.
Can Algorithms Be Considered to Have “Agreed” Under Section 3(3)?
Indian Courts understand that Cartels are essentially hidden and that agreements can be assumed based on the actions and circumstances of the parties involved. The Supreme Court in the case of Excel Crop Care Limited v. CCI2 has confirmed that the cumulative evidence of circumstances can be used to establish an agreement under Section 3; however, the Courts have cautioned against treating parallel behaviour as being equal to collusion, as illustrated by the Rajasthan Cylinders case3 which established that both parties must provide “plus factors” to indicate collusion.
Algorithmic Collusion disrupts this equilibrium. Where companies utilise similar or compatible algorithms to set prices, monitor one another and react to competitors, sustained parallel pricing may be an indication of something more than rational interdependence, it may also be indicative of coordination through algorithms producing cartel-like effects without the use of quantitative methods of communication.
A common defence to Algorithmic Collusion is the assertion that there is an absence of human intent as the prices are determined by autonomous systems; however, this undermines the very reason for the existence of Section 3 of the Competition Act. Competition law attributes the conduct of the enterprise to the enterprise themselves and not the methods or tools that they have adopted in order to undertake business. Algorithms are an integral part of an enterprise’s structure and are established and executed by the enterprise with certain goals in mind. The enterprise that chooses to establish a system that has the potential to produce collusive behaviour and that is always likely to produce anti-competitive behaviour should not be exempt from liability for these results because the enterprise is not involved in the day-to-day operations of the system.
Indirectly, Indian law lends credence to this conclusion by establishing that in Fx Enterprise Solutions, liability arose from a system’s coordinated structure and functionality rather than from explicit communications.
The reasoning behind this approach is equally applicable to algorithmic pricing systems that operate on the basis of generic market signals.
It is therefore possible to take a principled approach to this issue by changing the focus of the analysis from the subjective intent of firms to their awareness or foreseeability of the circumstances in which the use of a particular algorithm may weaken competition and sustain supra-competitive prices.
Thus, if one firm knows or ought to know that its use of algorithmic pricing systems results in weakened competition, or sustains supra-competitive prices, that firm’s continued use of that pricing algorithm can be seen to amount to an “action in concert” as defined in Section 2(b), regardless of whether that firm intended to engage in an anti-competitive act.
There is ample support in comparative jurisprudence, as illustrated in Eturas UAB v. Lithuania, to hold firms liable for knowing of and acquiescing to the existence and operation of an anti-competitive system. Indian law is well positioned to adopt this reasoning and apply it without necessitating the creation of a new legal doctrine or framework.
Conclusion
The rise of algorithmic pricing has created the need to rethink how to enforce cartel provisions in Section 3 of the Competition Act, 2002. The law was originally created to address agreements between people, but algorithms create new dimensions of illegal conduct because they can create cartels based solely on the interaction of algorithms.
Firms should not be able to say that it is legal or acceptable to engage in algorithmic collusion, just because it has been made through an algorithm. Section 3(3) does not preclude the use of circumstantial evidence to infer the existence of cartel-type behaviour, therefore, there should be no reason to prevent the making of algorithms to facilitate collusive activity.
Further, since these algorithms function as “autonomous” systems, firms cannot use the argument of the delegation of authority to avoid liability when it is foreseeable that the actions of these autonomous systems will result in sustained anti-competitive damage.
At the same time, enforcement must remain principled. An effects-based interpretative approach focusing on market outcomes, foreseeability, and control over algorithm design offers a balanced solution. As markets become increasingly governed by code rather than communication, competition law must evolve to scrutinise the competitive implications of algorithms themselves.
By entering the email address you agree to our Privacy Policy.