Algorithmic Memory ABSTRACT VANSH RASTOGI

Algorithmic Memory and Copyright Liability: Rethinking Infringement in Generative AI Systems

The emergence of Generative Artificial Intelligence technologies has led to an unprecedented development from a legal standpoint known as algorithmic memory, which is the ability of LLMs to store snippets of copyrighted content that the algorithms have come across in their training process in memory. Unlike traditional copyright infringements, which occur through human authors, algorithmic memory emerges secretly through model weights, and only surfaces later in the form of output.

This paper highlights two distinct areas where there could be potential liability - the training process and the output generation process. Both of these aspects are analyzed based on the context of increasing litigation cases in the international arena as well as domestically. This research highlights some cases including New York Times Co. v. Open AI, Getty Images v. Stability AI, Penguin Random House v. Open AI (2026), and the landmark case in India ANI Media Pvt. Ltd v. Open AI Inc & Anr., the world’s first judicial ruling in the case of copyright violation by generative artificial intelligence before the Delhi High Court, to demonstrate that the basic notion of human creativity inherent within the very structure of copyright law, namely the Copyright Act, 1957, and Digital Personal Data Protection Act,2023, which rendered obsolete in the age of artificial intelligence.

Given the policy themes of digital sovereignty and international equality that have been prevailing within the deliberations at the conference, this paper recommends the implementation of a multilevel liability framework and memorization auditing, compulsory licensing of training datasets and sui generis protection of artificial intelligence, in conjunction with WIPO. The notion of algorithmic memory requires a radical re-conceptualization of copyright law for the age of intelligence.

KEYWORDS: Algorithmic Memory, Copyright Liability, Generative AI Governance, Sui Generis Intellectual Property