INTRODUCTION: The Artificial Intelligence (Ethics and Accountability) Bill, 2025 was introduced as a Private Member’s Bill with the stated objective “to establish an Ethics and Accountability Framework for the use of Artificial Intelligence (AI) technologies in decision-making, surveillance, and algorithmic systems to prevent misuse and ensure fairness, transparency, and accountability and for matters connected therewith or incidental thereto”. The Bill seeks to serve as a general-purpose legislation to anchor AI-related developments in India. Given its sector agnostic stance, the Bill marks a departure from Government of India’s current regulatory approach which seeks to build on existing sectoral regulations rather than introduce a comprehensive overarching legislation. Adding to this context, we must also keep in mind the legislative history, where private member Bills in India have seldom been able to garner the necessary mandate at the floor of the house to become enforceable legislations. Given the drift in regulatory constraints coupled with prior history, it remains to be seen how much traction this Bill would be able to attract to become a law.

OVERVIEW OF THE BILL:
1. Centralised Governance by a singular authority The Bill proposes a novel solution to AI governance by proposing a single polycentric body comprising of members from academia, civil society and the government who may be experts in disciplines such as law, data science and human rights. This body, namely the Ethics Committee for Artificial Intelligence, shall be responsible for policy-guidance, regulatory actions and recommending penalties – including hefty monetary sanctions up to Rs. 5 crores and suspension or revocation of licences to deploy AI systems. Notably, the same body would be responsible for administrative oversight, policy direction and recommending penalties. Consequently, AI governance becomes highly centralised in the hands of a single organisation. Such a setup is directly in conflict with the governance architecture proposed by India AI Guidelines -which proposes a three-tier organisational structure that includes: 1. AI Governance Group – Serves as the apex body for enabling inter-ministerial coordination and administrative leadership at the highest level.

2. Technology & Policy Expert Committee – Advisory body consisting of academia, government as well as industry representatives to provide counsel and recommendations to the AI Governance Group. NEWSLETTER DECEMBER 2025 13 3. AI Safety Institute – Responsible for building and adopting technical standards for safe and responsible use of AI. Arguably, the India AI Guidelines provided a more comprehensive approach towards AI governance – by separately catering to building of technical standards, ensuring multi sectoral collaboration and coordination at the apex level – as compared to the proposed Bill. Moreover, no guidance has been provided by the Bill regarding how penalties, if imposed, maybe further contested. This lack of clarity may disproportionately impact startups and MSME’s as against larger enterprises, given that compliance risk-management and related burdens can potentially increase. Additionally, there is also no expressed opportunity to be heard provided within the Bill, before the Ethics Committee proceeds to impose sanctions. 2.Special Provisions concerning surveillance The Bill provides for special provision through Clause 5 for certain use-cases of AI-enabled systems. It provides that surveillance shall be limited to lawful purposes and would be subject to approval of Ethics Committee for Artificial Intelligence (AI). Unfortunately, the Bill does not define the terms “surveillance” or what would “lawful purposes” be construed as. Moreover, perhaps more critically, the Bill does not differentiate between surveillance functions performed by public bodies, such as Ministry of Home at the central level or Home Departments at the state-level, and private entities, such as banks, private lending institutions etc. Consequently, in the event of the Bill coming into force, key security agencies may have to apply for approvals to the Ethics Committee before deploying AI-enabled systems for purposes of internal security or defence purposes. Whether this has been intended deliberately or is an oversight remains unclear.

3. Regulation of AI systems involved in critical decision making Another provision in Clause 5 of the Bill deals with “AI systems involved in critical decision-making”. Such AI systems, convening functions that include but are not limited to law enforcement, financial credit and employment, are subjected to [a]what the Bill calls “stringent ethical reviews” and [b] carry the obligation to not discriminate based on race, religion or gender. Unfortunately, the Bill does not provide any objective standard to classify AI systems involved in critical decision-making. Given the same, there remains uncertainty as to which AI systems would qualify for being regulated under this clause. Moreover, what standard/threshold of inquiry constitutes “stringent” ethical review has not been clarified by the Bill itself. The phrase by itself carries the tendency to not only create uncertainty but also a chilling effect on startups and bootstrapped enterprises hoping to leverage from AI-related developments. Crucially, since usage of AI-enabled systems in hiring – both with within the public as well as the private sector may be expected to increase, there may be associated compliance requirements that can arise if the Bill is passed by both the Houses of Parliament that not DECEMBER 2025 14 NEWSLETTER only involve compliance on behalf of HR departments of corporate entities using automated systems to scan profiles but also constitutional bodies such as Union as well as State Public Service Commissions and other government recruitment agencies; besides added compliance responsibility on agencies in the internal security establishment as well as financial institutions.

4.Obligations on Developers of AI models and systems The Bill provides a definition for developers and imposes specified obligations on them. The term “developer” has been defined to mean “a person who design(s), develop(s), and implement(s) AI models and systems” as under Clause 2(b) of the Bill. Further, the developer is subjected to three obligations under Clause 6: 1. Obligation to ensure transparency: This requires the developer to disclose “intended purpose” and “limitations” of AI systems. 2. Obligation to prevent Algorithmic Bias: This obligation requires the developers carry regular audits to mitigate bias and ensure “diversity and inclusivity” in training datasets. The term algorithmic bias has been defined under the legislation to mean “presence of systematic errors in AI systems that result in unfair outcomes”. 3. Obligation to maintain records for compliance with ethical standards. In terms of its applicability, the definition of developer is agnostic to the nature and scale of AI models and systems, thus treating teams working for multi-national organisations at the same pedestal as school or college-going students incubating a startup. A graded approach to risk management depending upon nature and scale of AI based innovations would have necessarily ensured that the governance framework meets requirements of proportionality without stifling innovation. Moreover, developers as per the Bill are defined as those who “design, develop and implement” – such a framing is unfortunately misaligned with how AI-enabled tools are developed and deployed in the current eco-system. In practice, developing and deploying AI-enabled tools encompasses a wide range of entities including, though not always limited to, developers, vendors, entities that curate training datasets, cloud-service operators and end-users. Also, developers whose AI models are used for further building applications may not always have the requisite degree of control over their innovation; and risk attracting responsibility for consequences beyond their control. Moreover, there may be a need to distinguish between those engaged in developing foundational AI models and others who integrate these foundational models into sector-specific applications downstream. Unfortunately, the wide generalisations within the Bill fail to comprehensively account for such nuances. Additionally, the mandate for ensuring diversity and inclusion is imposed on “Every developer” under Clause 6. However, not all AI systems may in practice require “diversity and inclusivity” in datasets. For instance, AI models built for particular use cases such as handling network traffic or AI models being used to detect genetic anomalies in vulnerable communities may require targeted datasets to train models for accuracy rather than NEWSLETTER DECEMBER 2025 15 datasets with wide demographic diversity. Thus, adopting a blanket mandate for diversity and inclusion may not benefit all use-cases for AI-enabled systems.

CONCLUSION: A key feature of the Bill is its preference for broad definitions and general provisions, which if enforced would carry two essential imperatives. Firstly, they shall cast a wide regulatory dragnet on the AI innovation, which may be subjected to “stringent” ethical reviews as well as hefty penalties, for non-compliance. Secondly, the wide regulatory dragnet is accompanied by generic phraseology which may carry with it risks of overregulation as well as uncertainty within the AI ecosystem. While the proposed Bill pursues a laudable aim to lay down a comprehensive framework for ethical deployment of AI, its provisions often fall short of comprehensively dealing with issues that it, rightly, flags. The Bill, by hoping to be the anchoring legislation for AI governance in India, remains starkly in contrast with the regulatory direction which India-AI Guidelines and the RBI FREE-AI Framework has adopted in the past. Unlike the latter two, which pursue a sector-specific approach in dealing with AI-related developments, this proposed Bill aims to provide a comprehensive legislative approach to AI governance. While the jury is still out on which governance roadmap must India pursue, the Bill presents an opportunity to discuss future directions concerning AI governance – both within the legislative corri.