AI Agents as Autonomous Economic Actors: Who Bears Competition Law Liability When an AI Agent Negotiates, Prices or Purchases on Behalf of a Firm?

Author

Aditya Mishra, Ankul Prajapati, & Md Saqib Ansari

Abstract

 Autonomous AI agents increasingly negotiate, price and purchase on behalf of firms, yet competition law attributes conduct to humans and enterprises through doctrines of agreement, intent and concerted action. This paper examines the attribution gap that arises when self learning agents produce commercially significant, potentially anti competitive outcomes without clear human consensus. Focusing on the Competition Act, 2002, it argues that liability should remain with the deploying firm under an enterprise risk model, supplemented by a Control, Authorisation, Duty, Causation and Rule (CADCR) test to allocate responsibility across developers, vendors and beneficiaries where autonomy is high. Doctrinal analysis shows that Sections 2(b) and 3 can cover many AI mediated scenarios, especially in hub and spoke or common platform cases, but purely autonomous convergence strains the “meeting of minds” paradigm. Drawing on global enforcement trends in the United States, European Union, United Kingdom and Brazil, the paper proposes de lege ferenda reforms, including a targeted provision prohibiting algorithmic pricing systems that cause appreciable adverse effects, with rebuttable presumptions and disclosure powers. A compliance architecture is outlined, emphasising governance, algorithmic audits, data segregation and evidence preservation. The paper concludes that autonomy redistributes but does not eliminate accountability, and that Indian competition law should evolve from a purely consensus based model to an attribution framework tied to control, foreseeability and enterprise benefit.

Keywords

AI agents; competition law; algorithmic pricing; attribution of liability; enterprise risk; hub and spoke; Competition Act, 2002; CADCR test.

Full Text:

Download Paper PDF

References

The rise of AI agents as autonomous economic actors exposes a structural mismatch between competition law’s human centred doctrines and the realities of algorithmic markets. As this paper has shown, the core challenge is not whether AI mediated conduct can harm competition, but how to attribute that conduct to legal persons in a manner that preserves deterrence without stifling innovation. Under the Competition Act, 2002, Sections 2(b) and 3 provide sufficient breadth to capture many AI facilitated scenarios, particularly where common platforms, shared data feeds or hub and spoke arrangements enable coordination. Yet purely autonomous convergence by self learning agents strains the “meeting of minds” paradigm and risks an enforcement gap where harmful effects exist but no human consensus is provable.

The proposed solution is an enterprise risk and attribution framework. Primary liability should rest with the deploying firm, which chooses to use the agent for commercial decisions and reaps the benefits. The CADCR test offers adjudicators a structured method to allocate secondary liability across developers, vendors and beneficiaries based on control, authorisation, compliance duties and causal contribution. De lege ferenda, a targeted provision such as Section 3A, coupled with disclosure powers and mandatory algorithmic audits for high risk use cases, would align Indian law with emerging global best practice.

Ultimately, autonomy redistributes but does not eliminate accountability. Indian competition law should evolve from a purely consensus based model to one that ties liability to control, foreseeability and enterprise benefit, ensuring that firms cannot outsource compliance to code while still enabling responsible innovation in commercial AI.

Share your valuable work from Social Media Buttons