[By Saksham Agrawal]
The author is a student of National Law School of India University, Bengaluru.
Introduction
The CCI’s 2025 Market Study on Artificial Intelligence and Competition (‘2025 Market Study’) marks a decisive moment in India’s evolving engagement with digital markets. For the first time, it confronts not merely the deployment of artificial intelligence (‘AI’) as a business tool but its emergence as a participant in market coordination itself.
Pricing algorithms, which were previously appreciated for their efficiency and quick respons es , now act on their own as autonomous economic agents that can create results similar to, and sometimes better than, human collusion. The implications are profound because when coordination no longer requires communication, traditional antitrust concepts begin to fray.
The central question that follows is both conceptual and institutional. Can a legal framework built upon human intent, consensus, and “meeting of minds” extend to a world where coordination is computational?
This article examines how the 2025 Market Study reframes this problem within India’s competition law regime. It argues that while the study expands regulatory consciousness of algorithmic risks and proposes mechanisms for internal accountability, the Competition Act, in its present form, remains constrained by doctrines that presuppose human agency. The resulting gap is not of enforcement capacity but a gap between how markets now behave and how the law still thinks. However, a purposive interpretation of its wide wording to increase accountability among enterprises deploying pricing algorithms, along with building capacity through technical literacy, may provide solutions to the same.
Background
Algorithmic collusion refers to the coordination of prices or market conduct achieved through artificial intelligence systems, whether by deliberate human programming or through autonomous machine learning. What distinguishes these mechanisms from conventional cartels is not the outcome they produce, but the process by which they reach it.
The Study refers to 4 identified operational forms. These are monitoring algorithms, parallel algorithms, signalling algorithms, and self-learning algorithms.
Monitoring algorithms serve as instruments of execution or surveillance, implementing or monitoring agreements that have been consciously formed by human actors. Parallel algorithms or Hub-and-Spoke algorithms involve competitors relying on a common pricing algorithm or platform, enabling indirect coordination of prices through a shared technological intermediary. Signalling algorithms capture situations where firms independently deploy reactive algorithms that adjust to market conditions in similar ways, thereby increasing the likelihood of tacit alignment without explicit agreement. Finally, self-learning algorithms autonomously optimise prices through iterative learning, giving rise to collusive outcomes that occur without human intent, awareness, or participation.
The first three categories fit comfortably within the conceptual apparatus of antitrust enforcement: they depend, at some level, on human design or tacit consensus. The fourth does not. It involves coordination without communication.
The resulting difficulty is not merely evidentiary but ontological. Competition law has always treated collusion as an act of will, a convergence of minds expressed through behaviour. But when algorithms learn to align independently, what remains of collusion once its human authors disappear?
The Competition Act’s framework
Section 3 of the Competition Act prohibits agreements, decisions, and “actions in concert” that cause or are likely to cause an appreciable adverse effect on competition. Its open-textured phrasing reflects deliberate legislative breadth where it captures conduct that may fall short of an explicit contract but nonetheless reveals a common economic design.
This elasticity offers interpretive room to address algorithmic coordination. Where enterprises knowingly deploy similar pricing algorithms, or adopt machine-learning systems calibrated to react to market data in comparable ways, the resulting interdependence may constitute an action in concert even in the absence of explicit communication. In theory, then, the statutory language is capacious enough to include technologically mediated coordination within its fold.
Yet the structure of Section 3 remains rooted in an anthropocentric model of collusion. It assumes actors capable of intention, decision, and mutual awareness, all attributes that belong to legal persons, not autonomous systems. Self-learning algorithms fracture this premise. They act without instruction, evolve without oversight, and generate patterns of market alignment that no individual firm may have foreseen or even understood.
The law presupposes agency as a precondition for culpability, but in algorithmic markets, agency is diffuse, distributed between code, design, and data. The result is a conceptual disjunction between how the law identifies responsibility and how coordination now occurs.
The 2025 Market Study implies that accountability should remain with the enterprise deploying the system, but the Act provides no clear doctrinal bridge between control and outcome once human intervention ceases.
If liability follows control, firms could evade responsibility by distancing themselves from their algorithms’ autonomy. If it follows effect, firms risk sanctions for conduct they neither intended nor could reasonably predict. The absence of an intermediate principle that ties responsibility to foreseeability and design risks creating a zone of regulatory paralysis precisely where control has been surrendered to machines.
The Market Study’s Approach
The 2025 Market Study acknowledges that algorithmic interaction can generate and sustain supra-competitive pricing even in markets that are neither concentrated nor consciously collusive. What distinguishes such outcomes is not conspiracy but code per se and the capacity of algorithms to observe, infer, and adjust with a speed and precision far beyond human coordination.
The study identifies three structural attributes that make algorithmic collusion uniquely resilient. First is its speed, which compresses the interval between detection and retaliation; second, opacity, which obscures causation and intent; and third, interdependence, which ensures that one system’s decision becomes another’s signal. Together, they create a feedback loop that can stabilise collusive equilibria without any act of human agreement.
In response, the study proposes a shift from reactive enforcement to preventive compliance. It recommends that enterprises conduct algorithmic self-audits consisting of systematic reviews of how their AI tools function in practice, documenting design parameters, data inputs, and market outcomes. Firms are encouraged to test algorithms periodically to detect patterns of unintended coordination and to maintain internal documentation explaining how pricing decisions are reached.
A second strand of reform concerns hub accountability. Digital platforms and intermediaries that deploy common algorithms for multiple market participants may act, even inadvertently, as nodes of coordination. The study suggests that such entities bear a heightened duty of vigilance, ensuring that their technological infrastructure does not facilitate anti-competitive alignment.
I believe this evolution unsettles established safeguards. If liability turns on outcome rather than intention, questions of fairness, predictability, and proof inevitably arise. How should regulators distinguish between legitimate algorithmic adaptation and illicit alignment? How can enterprises demonstrate due diligence when causation itself is opaque? The study raises these questions without fully resolving them, leaving the architecture of accountability still under construction. This problem of algorithmic collusion is not new; it has been identified previously (see here and here). Thus, I believe these questions need to be resolved better.
Self-learning algorithms represent the point at which competition law confronts its conceptual horizon. These systems do not execute pre-programmed strategies; they discover them. Through continuous observation and reinforcement, they can autonomously infer that cooperative pricing maximises profits and in doing so, arrive at a stable equilibrium that mimics deliberate collusion.
This is coordination achieved not by exchange of wills, but by iterative logic. Traditional doctrine is unprepared for such behaviour because it rests on human intentionality, like an agreement, a concerted practice or a conscious alignment. Where the algorithm acts alone, that architecture collapses.
For enforcement, the implications are immediate. No agreement means there is no direct evidence to seize upon. Then, no intent means there is no culpability in the classical sense. However, the effect of prices that rise and remain above competitive levels is functionally indistinguishable from a cartel.
The 2025 Market Study concedes as much. It recognises that this “digital-eye” coordination, i.e., autonomous learning that yields collusive outcomes, lies beyond the present reach of enforcement. The gap is not simply procedural. Competition law is designed to discipline choice, not consequence. When algorithms generate anti-competitive effects without human direction, the law finds itself stranded between logic and liability, aware of harm but unable to ascribe blame.
Reinterpreting liability and redesigning governance?
There are 2 possible responses to this problem. First, doctrinally, section 3 of the Competition Act possesses the conceptual elasticity needed to adapt to algorithmic markets if interpreted purposively. The phrase “action in concert” can, and should, be understood to include algorithmic conduct that firms knew or ought reasonably to have known would align with competitors’ systems. This would anchor liability not in subjective intent but in foreseeability and the recognition that enterprises deploying adaptive algorithms bear responsibility for anticipating how their systems interact in the marketplace.
Such an approach shifts the inquiry from intent to risk. It treats algorithmic coordination as a matter of accountability by design, where culpability arises not from direct communication but from the failure to foresee and prevent foreseeable collusive effects. The alignment of liability with knowledge and control rather than mere intent , allows competition law to broaden its application without compromising its core principles.
The second response is what can be best termed a policy response by embedding algorithmic transparency and auditability within the framework of competition compliance. Enterprises deploying pricing algorithms should be required to maintain detailed documentation of design logic, data sources, and performance outcomes, subject to periodic verification instead of being merely advised to self-audit.
Rather than creating a separate regime for AI ethics, these obligations can be integrated into the Competition Act’s existing compliance architecture under Section 49 guidance. A coherent regulatory ecosystem would also demand inter-agency collaboration, particularly between the CCI, MeitY, and emerging AI oversight bodies (see here and here) to develop uniform technical standards for algorithmic auditing and accountability. Such coordination would prevent regulatory fragmentation while ensuring that enforcement keeps pace with technological sophistication.
Now, the success of this transition will depend less on expanding the CCI’s statutory mandate and more on deepening its technical literacy. Algorithmic coordination cannot be inferred through traditional economic analysis alone. It requires AI forensics, data analytics, and market simulation to uncover the behavioural logic embedded in code. These techniques enable the CCI to move beyond static price analysis and instead understand how algorithms interact dynamically and how patterns of adjustment, response, and learning can transform competitive markets into coordinated ones without explicit human direction.
Building this capacity is neither cosmetic nor optional. Without the expertise to interpret algorithmic decision-making, even the most progressive policy framework risks remaining operationally inert. The future of competition enforcement will therefore turn on the CCI’s ability to transform itself from a market regulator into a technological diagnostician, capable of reading not just what firms decide, but how their machines learn.
The road ahead
The 2025 Market Study signifies more than a policy exercise. It marks the transition from conceptual awareness to institutional readiness. The regulator now recognises that the frontier of competition law no longer lies in identifying agreements but in understanding interactions itself where algorithms, not executives, may determine the boundaries of competition itself.
The immediate objective is to incorporate accountability into the design process, rather than simply identifying collusion through its outcomes. Compliance must begin in the architecture of algorithmic systems, in the parameters that govern their learning, and in the oversight mechanisms that monitor their adaptation.
The language of the Competition Act is flexible enough to evolve, and what it requires is purposive interpretation rather than textual literalism. Enterprises cannot escape responsibility through the opacity of their algorithms. The law’s duty is to ensure that technological autonomy does not translate into regulatory absence. In the years ahead, the measure of India’s competition regime will lie not in how swiftly it adapts to technology, but in how faithfully it preserves the idea that even in digital markets, accountability must always have an author.