By – Aashita Sarin
Abstract
Imagine a citizen’s information is legally posted in one instance but then continues to flow within the AI system long after it is legally and socially no longer relevant and there is technically no way of removing it. This is not an unusual scenario. It is the state of present-day large language models where information is spread out throughout billions of parameters in a way that can’t be found, isolated, or deleted without tampering with the model. This article contends that the clash between the right to be forgotten and the nature of contemporary AI is not temporary but a structural one and that the consequences are likely to impact the most vulnerable in society, those whose history has the greatest potential to become problematic. Based on the European law on data, the principles of Indian Constitution and the dharmic concept of identity as constantly re-created and not accumulated, the article contends that India has a limited window for the purposes of accommodating forgetting as a design criterion, failing which the infrastructure will do it for them.
Introduction
It is the law that will decide who will get mercy and who will not. The law has always known that without mercy, there is no justice. The criminal records are closed. Juvenile proceedings are kept confidential. Bankruptcy filings expire. The following disabilities are not merely administrative matters or symbolic concessions. The legal system already knows that accountability has a shelf life. A record that lingers long after its relevance has expired no longer serves justice.
The instinct has philosophical antecedents older than any statute. In the Upanishads, the agential conception of personhood suggests that a person’s identity is shaped by their actions, and karma provides the possibility for renewal. Personhood is not a ledger. It’s a process. When village panchayats chose to set aside certain histories, they were not being imprecise. They were making a moral judgment about which version of a person deserved to persist, since letting every record persist indefinitely is not neutrality. It is a choice.
The right to be forgotten must be understood in this philosophical and legal context, which is not merely technical or one-off but has a long pedigree: the principle that identity is not eternal, that relevance is not eternal, and that a proper society does not fix its citizens forever by the worst moments of their lives or the most troublesome. This principle was legally established in 2014 by a ruling of the Court of Justice of the European Union, which ruled that when a consumer has settled a debt, an old newspaper ad must be removed from the results of a search. India has arrived at the same position, using the reasoning of Justice Kaul in Puttaswamy v. Union of India (2017), that the right to be forgotten is part of the constitutional right to privacy under Article 21 and, in a more speculative sense, the Digital Personal Data Protection Act of 2023. Both these theories couldn’t have predicted a technology that not only retains but also encodes and propagates information and that makes it immune to the idea of erasure.
Two traditions, one conclusion.
Europe created its framework by means of legislation. India had its own reasons. Such a culture, which had always known identity as mobile, contextual, and not reducible to the recorded, did not have to borrow the argument. Justice Kaul’s acknowledgment of the right to be forgotten as a part of informational privacy under Article 21 was based on constitutional dignity, which Indian jurisprudence had been quietly evolving all along. What has not followed is the legal architecture to protect that understanding within systems that were not built to forget.
The Architecture of Permanent Memory
Search engines index pointers, and pointers can be removed. That is what GDPR Article 17 was built around, and for that architecture, it worked. Large language models are a different problem entirely. Information is not stored in retrievable chunks in training. It is distributed in hundreds of millions of parameters throughout the entire network, entangled with everything else the system has learned. There is no address for one fact. No row to delete. The problem is not trivial. It is structural, erasure requires locating something not stored as a locatable thing. The law presumes an architecture that does not exist within these systems.
One such field that has emerged as a solution to this is called “machine unlearning.” This is the field trying to work around the problem at hand. None of these approaches has yet reached the precision required by erasure law. When the industry says compliance is too expensive it is not describing an impossibility. It is describing a decision. The cost doesn’t go away. It is passed on to the very people the law was designed to protect. That’s not a technical limitation. It is political.
Who Actually Pays
This debate tends to be abstract. But the reality is not. The people most impacted by permanent machine memory are not public figures managing their reputations. They are the person named in a case file they never wrote. The survivor whose disclosure in what she believed was a safe space was scraped years later into a training corpus. The witness who appears in a system they never signed up for and can’t get out of. These are not exceptions. They are the cost of making erasure optional. The financial hardship of a sick person has become a part of an employment background check with no way to present the circumstances that would ensure the information is not only condemning but also understandable.
These individuals don’t consider the system’s lack of forgetfulness to be a nuisance. It is a process that hinders what they want to do at present due to what they failed to do in earlier times. The panchayat elder who didn’t raise a certain history was doing something the algorithm structurally can’t do: applying their judgment on what it costs the remembered person to be remembered in that manner. Optimizing language patterns does not yield such an evaluation of the moral issue.
The Window India has
Europe is trying to fit the square peg in the round hole of privacy law. The place is in a more nascent stage, and in some ways, that is uncomfortable, but in others, that’s a great thing about India. The DPDPA 2023 exists but the secondary legislation is still being drafted. The question of how deletion rights apply to AI systems is still unanswered. That gap is no problem. It’s an opportunity. A serious framework would reverse the logic: companies would need to justify having personal data, and the cost of erasure would be a cost of doing business, not an excuse for non-compliance. This is not a radical stance. That is what both GDPR and Indian constitutional privacy jurisprudence already point to.
Conclusion
A newspaper archive can be petitioned. A search result can be removed. A large language model cannot be asked to forget in any equivalent sense. Information inside it is not stored in a location. It is a weight, an association woven into the way the system thinks, i.e., distributed, entangled, and unreachable by any engineer with a delete key. In the philosophy that recognizes the fluidity of identity and its right to transcend the past, there is nothing like the architecture of a model that was never designed with mercy in mind. The infrastructure is still being built up. India has the philosophical basis, the constitutional precedent and a regulatory window that is not yet shut. The people waiting inside that window aren’t abstractions. They are the ones whose lives will determine whether the right to be forgotten will become enforceable law or a promise the machine was never built to keep.
About the author
Aashita Sarin is a second-year undergraduate psychology student at O.P. Jindal Global University. She is interested in artificial intelligence, large language models, and how technology influences the way people think, behave, and make decisions. Her work focuses on the ethical, cognitive, and social impact of AI, especially the role of intelligent systems in everyday life and public decision-making.