The Government of India has come up with AI Governance Guidelines (2025) as non-binding policy guidance. This paper proposes to give them more than advisory significance and to make them increasingly significant in the institutional conduct of institutions and in their governance. It analyses the constitutional implications of this change and the issues it raises for democratic accountability and legislative oversight.
I. Guidance and the Constitution Puzzle of Government
The fundamental principles of the Constitution are generally assumed to govern the exercise of public power. The Indian system of law is enacted by the Parliament, implemented by the Executive branch and checked for constitutionality by the Courts. The legitimacy of State action is not just about what action is taken, but also how it is taken. Yet algorithmic governance unsettles this constitutional set-up. In the regulation of AI systems, executive instructions are increasingly taking the place of legislation across welfare, taxation, digital identity, public procurement, regulatory enforcement and administration. The AI Governance Guidelines (2025) are, at first glance, policy documents, but are also important as tools for responsible innovation and risk-informed deployment.
This constitutional concern does not arise merely because the executive issues policy guidance. The worry is that these tools are increasingly stepping in to perform functions like those of constitutional ordering. They shape conceptions of accountability within public institutions, decision-making practices, the assignment of responsibility and what constitutes acceptable forms of algorithmic power. Administrative advice begins to function as a normative framework for state action.
II. The Production of Administrative Norms in the Context of Soft Law
In the classical theory of administrative law, there is a distinction between rules and instructions of the executive. The former derive authority from legislation, and impose obligations that are enforceable. The latter are typically used as internal administrative instructions, rather than as legal instruments.
The formal distinction is often challenged, however, in the context of modern regulatory states where soft law is at work. The effects of non-binding instruments are likely to be more effective in influencing institutional behaviour than formal legislation. The AI Governance Guidelines (2025) fall squarely within this phenomenon. This dynamic is particularly important as the Guidelines are not standalone. They are supported by recommendations for institutional frameworks like the AI Governance Group (AIGG), the AI Safety Institute (AISI) and the Technology and Policy Expert Committee (TPEC). These entities combine to create an ecosystem that can generate governance standards, technical recommendations and risk-management frameworks. The document also claims that existing laws (the IT Act, the DPDP Act, and sectoral laws) are already sufficient for the formal purpose of excluding the need for primary laws, thus placing executive soft law as the most important source of norms. This leads to the development of an administrative infrastructure with norm-generating functions which is not mediated by traditional legislative procedures. The execution of the rule-making function is concurrent, not sequential, as the Executive is creating both the institutions and the standards.
III. Constitutional Path Dependency and Algorithmic Governance
The constitutional implications of this development become clearer through the lens of path dependency. The technical standards established at an early stage in regulation can take hold in a variety of ways, including procurement contracts, compliance frameworks, institutional training and bureaucratic expectations. The AI Governance Guidelines (2025) contain several short- and medium-term implementation measures aimed at speeding up this process. Risk frameworks, incident-reporting mechanisms, audit processes, and voluntary commitments may initially appear experimental. However, once embedded in public administration, they begin influencing future regulatory decisions. The Action Plan included in these Guidelines focuses on the constitution of these bodies, the development of risk frameworks specific to India, and the establishment of mechanisms to report incidents – all of which speed up this embedding process.
Executive guidance does not become constitutional at this point, not because it is a formal amendment to the law, but because it sets the underlying assumptions for the evolution of the law in the future. Five years later, Parliament would face a governance ecosystem already in place through executive initiative before considering AI legislation. The real-world effect is subtle, but not insignificant. Executive preferences become institutional defaults, prior to public deliberation. In this way, the algorithms and algorithmic governance show us what soft law is and how it can become a constitutional reality without being constitutional law.
IV. Democratic Legitimacy and Executive Constitutionalism
Regulatory agility is one of the arguments adduced in defense of the soft-law model. The pace of change in AI is much faster than the pace of legislation. Executive leadership enables governments to deal with technological developments without delay, when a lengthy parliamentary process would take place. However, in a constitutional democracy, efficiency is not the only concern, it is also important to have legitimacy. The point is not whether executive institutions can keep up with emerging technologies’ regulation faster than legislatures, but should they have primary authority to set the normative limits of algorithmic power?
Standards issues of this type are not amenable to being considered as “technical questions”. All algorithmic frameworks carry values. The thresholds for errors, fairness measures, transparency requirements, audit limits and risk classifications naturally favour some interests over others. Public accountability is essential and cannot simply be assumed by administration. But the architecture designed by the Guidelines puts these questions more into the realm of executive and expert institutions. Technocratic coordination is prioritized, with public contestation taking a backseat. The outcome is likely to be what can be called “executive constitutionalism,” whereby the core principles of governance are devised largely by the executive and not by the people through discussion.
V. The Missing Constitutional Vocabulary
Another challenge relates to how algorithmic harms are interpreted in a given language. The Guidelines focus primarily on issues related to safety, innovation, trustworthiness, risk management and appropriate deployment. These goals are very significant but not necessarily synonymous with constitutional accountability. The questions asked by a constitutional framework are different. It asks whether automated systems produce arbitrary outcomes, whether individuals can meaningfully challenge algorithmic decisions, and whether public institutions can explain automated reasoning. It poses the question of whether the state is acting in an equitable and fair manner. The concerns are directly related to Articles 14, 19 and 21 of the Constitution.
The Guidelines do not speak in constitutional language, but rather take the approach of considering algorithmic harms to be technical or ethical problems of responsible innovation. If problems in algorithms are simply blamed on their technical aspects, the most common proposed remedies are better data, more audit, and better design of the system. This distinction is important because not all constitutional harms can be solved by the technical solution. Even if it is entirely accurate, a system can be problematic if it has the effect of concentrating power, lessening transparency or weakening democracy’s oversight.
VI. Who Writes the Rules of the Algorithmic State?
There is a constitutional challenge that is often overlooked in discussions on AI. Algorithms often produce decisions that are the subject of public debate. But an equally important question concerns the institutions that define the rules those algorithms follow. India’s AI Guidelines reflect the evolution of executive soft law towards constitutional significance. This challenge is more nuanced than an overt statement of power. Soft law allows for rudimentary governance decisions to be made without a full reckoning with their constitutional consequences. The threat is that, if it is not democratically enacted law, the architecture of AI becomes normalised through the practice of the executives.
In the algorithmic age the constitutional debate is thus not technical or even regulatory only. It is fundamentally institutional. The central question is who will be able to set the rules for the legal framework of the digital state as it moves towards some new forms of algorithmic governance. Where this occurs primarily through soft law, constitutional accountability risks becoming an afterthought rather than a foundation. A democratic republic cannot allow the rules governing public power to emerge outside the constitutional processes that give that power its legitimacy.
