Introduction

The present-day economy has transformed into a fully digitised landscape, characterised by algorithmic systems that continuously monitor competitive environments and recalibrate prices in real-time to attract the maximum number of consumers. As firms adopt dynamic pricing tools, machine-learning-based demand-forecasting models, and autonomous decision engines, competition law faces a profound conceptual crisis: when market outcomes resemble collusion but no human agreement exists, should liability still arise? This concern is no longer speculative. International bodies such as the OECD have time and again cautioned that self-learning algorithms may independently converge on collusive equilibria without any explicit conversation between the firms.        

The problem of algorithmic collusion is far from theoretical; it is unfolding in practice across digital markets. Online travel agencies, e-commerce platforms, and ride-hailing services already rely on algorithmic pricing systems that respond instantaneously to rivals’ behaviour, sometimes creating outcomes that mimic coordinated strategies. For instance, a notable empirical study of retail gasoline markets in Germany found that the adoption of algorithmic pricing software resulted in a significant increase of 28% in profit margins despite the absence of any express agreement between competitors. Although neither illegal nor intentionally anti-competitive, the study revealed how algorithmic autonomy can produce extreme, super-competitive and unforeseen outcomes even without human collusion.

As artificial intelligence (AI) systems grow more refined, the boundary between optimisation and collusion will continue to blur even further. The challenge for competition authorities worldwide is to determine whether—and how—legal frameworks should evolve to recognise forms of coordination that arise not from human intent, but from algorithmic interaction operating at speeds and complexities beyond human capacity.

Algorithmic Collusion beyond Traditional Cartels

Traditionally, collusion has been understood through the framework of cartels—agreements between human actors based on the classical elements of a contract: a meeting of minds, communication, clear intentions, and conscious coordination. Algorithmic collusion disrupts this foundation by enabling forms of parallel conduct that are stable, predictable, and mutually beneficial, yet often unintentional and emergent. Scholars such as Ariel Ezrachi and Maurice Stucke in their book have argued that algorithmic systems can create “virtual competition havens,” where machines interact in ways that humans neither supervise nor fully understand.

Algorithmic collusion typically arises through four models of interaction: messenger, hub-and-spoke, predictive, and autonomous. The first two resemble classical anti-competitive conduct, as firms knowingly deploy software to execute or monitor collusive arrangements. However, the challenges intensify with predictive and autonomous algorithms. Predictive systems analyse rivals’ past behaviour to forecast future pricing patterns, potentially enabling tacit coordination. Autonomous algorithms—especially those trained through reinforcement learning—may independently discover that collusive strategies maximise profits and stabilise them without ever receiving explicit human instructions.

This evolution raises a significant policy dilemma: if no human intent exists, can algorithms themselves be “blamed” for collusive outcomes? More critically, should liability extend to firms when such coordination is not a deliberate business strategy but an emergent property of opaque AI systems? These questions expose a growing gap between traditional competition law principles and the realities of machine-driven markets, underscoring the urgent need for frameworks that can account for algorithmic autonomy and its profound implications for enforcement.

Why Existing Competition Law Frameworks Are Ill-Equipped?

Not only in India, but across different jurisdictions, competition law frameworks remain rooted in the human agency. This is clearly reflected in the European Union’s approach to competition enforcement. The European Commission continues to rely on the requirement of an agreement or concerted practice grounded in communication and a meeting of minds, even while acknowledging the risks posed by algorithmic pricing. Similarly, U.S. antitrust enforcement under the Sherman Antitrust Act is premised on the existence of a conspiracy- an inherently human-centered standard. In India, the Competition Commission of India (CCI) has yet to articulate a clear position on autonomous algorithmic collusion, though its 2020 e-commerce market study flagged concerns regarding opaque pricing mechaniss and platform-driven co-ordination.

The legal frameworks start to strain when conscious parallelism occurs at a large scale.      Courts have long held the notion that, parallel conduct is not sufficient to establish collusion unless it is supplemented by “plus factors”- meaning behaviour which cannot be explained by each company acting alone. Autonomous algorithms, however, can replicate these plus factor-like behaviours without any human communication, thereby producing the cartel-like outcomes while still fitting neatly outside the current legal definitions.

The idea of intent- which is one of the core principles of competition law- becomes even more complex to apply with reinforcement-learning algorithms. These algorithms do not work on the fixed rules; they learn from their environment and adapt according to it. because of which, their decision making is so complex and often impossible to fully understand, it becomes difficult to assign culpability on anyone, in case of their actions. 

Economic Dynamics of Algorithmic Market Coordination

The economic logic behind algorithmic collusion comes from the way the AI systems predict and react to the changes in the environment. Unlike humans, they do not suffer from impatience, bias, mistakes due to emotions, or fear of being caught, because of this nature they never undervalue profits. These AI systems also work constantly, and respond within milliseconds to rivals’ price change. Economic studies show that when reinforcement-learning algorithms are used in repeated pricing situations, they often end up choosing collusive pricing strategies, because these strategies end up delivering the highest long- term profits.

A study by the University of Bologna and MIT has found out that even when no instructions to collude have been given to the independent working Q-learning algorithms they still ended up keeping prices higher than competitive levels. This challenges the popular belief that ‘Tacit Collusion’ only happens in markets where there are few dominant players. Instead, it shows that algorithms can create stable collusive outcomes even in markets that are too unstable for human-led cartels.

These algorithms also make markets transparent to a great extent. With real-time data tools, firms can within seconds, see when the competitor changes prices. In the past, delays and uncertainty made tacit collusion hard to maintain. But this constant visibility removes those barriers, meaning that while transparency can help consumers, it also increases the risks of firms settling into coordinated, higher-priced market behaviours.

Liability without Intent?

One of the most heated debates in the regime of competition law is whether law should impose liability in the absence of human intent. Some scholars argue for ‘strict liability’, basing their beliefs on the fact that firms deploying powerful AI tools should bear the responsibility for any anti-competitive outcome, intentional or otherwise. While some caution against this reasoning, based on the fact that it may stifle innovations, the small firms will suffer by this reasoning as they mostly rely on third-party pricing algorithms.

There is another group which posits for ‘foreseeability-based liability’, whereby the firms will only become accountable for the actions and risks which were reasonably foreseeable by the firm at the time of deploying these powerful software. However, foreseeability in itself is contested when dealing with non-deterministic AI models whose behaviour cannot be fully predicted at the time of deployment.

EU jurisprudence offers partial guidance. In Eturas, the Court held that even passive acceptance of system-generated communication may constitute participation in a cartel, suggesting that reliance on shared digital infrastructure triggers a heightened duty of diligence. While the facts in Eturas involved human-readable messages, the underlying principle hints at an emerging doctrine: firms cannot outsource potentially anti-competitive coordination to automated systems and then disclaim responsibility.

The Indian Context: Gaps, Latencies, and the Road Ahead

India’s competition law, based on the Competition Act, 2002, is strong when it comes to deal with traditional human formed cartels, but it does not directly address the algorithmic collusion present in the market. So far, the CCI has not handled a case involving fully autonomous pricing algorithms. However, the issues relating to algorithmic collusion have already appeared in investigations of e-commerce platforms, where machine-learning tools used for discounts, prices and product reviews have raised a serious doubt about anti-competitive behaviours.

The challenge related to algorithmic collusion is growing, as the market is becoming more digitised and is depending heavily on AI-driven fintech apps, ride-hailing services, and food delivery platforms. Since India does not yet have clear rules for algorithmic governance, it remains unclear on who will be accountable when multiple companies will use the same third-party pricing software.

Another major problem in detecting algorithmic collusion in India would be lack of knowledge and instruments to detect these within the enforcement bodies. For detecting the algorithmic collusion, understanding of code, analysing algorithmic behaviour, and separating normal optimisation from patterns that indicate collusion. With the lack of technical expertise and lack of skill-set, the enforcement may become slow, superficial or ineffective.

Regulatory Responses: Global Proposals and Emerging Models

Across the world, regulators are trying to make regulations to deal with algorithmic collusion. In the European Union (EU), the Digital Markets Act (DMA) has created rules for big “gatekeeper” platforms that indirectly reduce collusion risks by limiting self-preferencing and requiring more transparency. The European Commission has given regulators “audit-rights”, by which they can inspect high-risk algorithms, which are likely to become quite important in future enforcement.

In the United Kingdom (UK), the Competition and Markets Authority (CMA) promotes algorithmic audits, transparency requirements, and even structural changes to companies when needed. In the US, the Department of Justice has made it clear that companies cannot avoid liability just by blaming their pricing algorithms—if the software helps them coordinate prices, they will still be held responsible.

Even though these regulatory efforts differ from country to country, a shared understanding is emerging that regulators need access to algorithmic systems, companies must closely monitor and understand the AI tools they use, and competition law must adapt to address new, machine-driven forms of coordination.

Reimagining Competition Law for the Algorithmic Era

Future-proofing competition law requires a complete shift in the concepts and reforms. Firstly, the definition of “agreement” should include within itself coordination mediated by the learning algorithms, even without explicit human communication. Secondly, the regulators shall impose mandatory transparency and auditability for high-risk AI systems, particularly in oligopolistic markets where risk of collusion is significant.

Thirdly, there is an imminent need for the adoption of algorithmic compliance frameworks, requiring firms to monitor and document the behaviour of the AI system deployed by them, conduct stress tests on them to detect collusive tendencies, and report the anomalies to the regulators. Such proactive measures may help distinguish between legitimate optimisation from anti-competitive behaviour.

Finally, competition law must integrate within itself interdisciplinary knowledge from fields like computer science, behavioural economics, and data governance, to address the complexities of machine economy. As the market is becoming digitised, the intellectual architecture of competition law must adapt accordingly.

Conclusion

Algorithmic collusion represents a fundamental challenge to the foundational assumptions of competition law. AI systems, operating at speed, scale, and sophistication far beyond human capacity, can create market conditions that not only mimic but potentially surpass the stability of human cartels. Traditional doctrines—rooted in intent, communication, and conscious coordination—are no longer adequate. As jurisdictions worldwide grapple with these issues, the path forward lies in embracing transparency, auditability, cross-disciplinary collaboration, and a willingness to reconceptualise liability in an era where algorithms behave as autonomous market agents. For India, this moment presents an opportunity to craft forward-looking regulations that balance innovation with competitive integrity, ensuring that AI remains a force for efficiency rather than a quiet architect of collusive harm.

About the Author

Sakshi Upadhyay and Utkarsh Utsav are students at the Chanakya National Law University, Patna.