Algorithmic Accountability: A Legal Framework for Board Oversight of Agentic AI Systems in India

Artificial intelligence is rapidly advancing and changing from being a passive partner in generative AI to developing into agentic systems enter into contracts, and carry out real-time financial operations with little human direction. As these agentic systems become embraced by more of India’s large businesses and incorporated into their normal operations, there has been significant development of an "accountability gap" in the traditional corporate governance framework. This research has revealed a significant conflict between the autonomous nature of agentic AI and the fiduciary duty of directors to act with care as described in Section 166 of the Companies Act. Specifically, if an AI agent carries out a transaction that violates either of the market norms or privacy laws, the traditional "principal-agent" relationship outlined in the Indian Contract Act of 1872 will not apply, as the AI agent possesses no legal personality or "sound mind". Additionally, this paper also proposes that the doctrine of "reasoned decisions" in administrative law should also apply to corporation boards, as directors cannot perform their duty of care without relying on algorithms that provide no rationale for how they arrived at their conclusion when there is no way to explain their outputs and provide an explanation. This paper aims to develop an innovative multilayered algorithmic accountability legal framework using data obtained from comparative analysis of SEBI advisories with other global artificial intelligence governance-related principles. The proposed legal structure contains provisions for algorithm governance through design; establishment of board level AI ethics committee; Finally, this paper proposes that in order for India to be recognised as a leader in the digital economy, it must extend its current legal/regulatory framework, to ensure that while AI will be able to operate independently of human input, the ultimate responsibility for all consequences of AI's independent actions must be held by human decision makers.