Requiring AI Companies To Obtain Multiple Licences For LLM Training Would Be Economically Unviable: Delhi High Court

The Delhi High Court has observed that requiring artificial intelligence (AI) companies to obtain licences from multiple sources to train Large Language Models (LLMs) would make their development economically unviable, while refusing to grant an interim injunction sought by ANI Media against OpenAI in its copyright infringement suit over ChatGPT.

"The key to success of an AI model is to access the information in public domain. The development of LLMs and their success depends on availability of data. It would be economically unviable to develop an LLM if training of an LLM would require licenses from multiple sources," Justice Amit Bansal observed.

The observation was made while deciding ANI's plea for an interim injunction seeking to restrain OpenAI from allegedly using ANI's copyrighted news content to train ChatGPT's underlying Large Language Models (LLMs) and from reproducing ANI's copyrighted works in ChatGPT's outputs.

The court made the observation while weighing the balance of convenience and public interest in deciding whether ANI had made out a case for interim relief.

The court noted that generative AI has brought about a transformational change across the world, including in India, and has revolutionised the manner in which people seek and obtain information.

Referring to NITI Aayog's AI for Inclusive Societal Development paper, the court observed that AI applications are being used beneficially in sectors such as education, healthcare, the financial support sector, agriculture, and skill development.

Building on this, the court observed that an interim injunction at this stage would be "detrimental to the growth of AI and more particularly, to the LLMs being developed in India.

"It would also have adverse impact on public interest, including millions of users of ChatGPT in India, many of whom would not be paid subscribers. Public interest is an important aspect or the fourth factor that the Courts have to consider while granting interim injunction, more particularly in intellectual property cases," the court observed.

The court also noted that ANI had offered OpenAI a licence for its content for a fee of USD 7.5 million through a communication dated October 3, 2024. The court observed that this prima facie indicated that ANI's claim was capable of being compensated monetarily if it ultimately succeeded in the suit.

By contrast, the court observed that an injunction against OpenAI would significantly affect its functioning and the resulting harm may not be capable of monetary compensation.

While considering the balance of convenience, the court noted that ANI could have blocked OpenAI's web crawlers from accessing its website through the available opt-out mechanism but had not done so.

It also recorded OpenAI's submission that it had blocked ANI's website from being scraped for training and from ChatGPT Search's Retrieval-Augmented Generation (RAG) feature. The court added that if ANI's subscribers had not blocked access to their websites, it may have been for commercial reasons.

The court also noted that ANI had not placed any material on record to show that OpenAI's activities had caused it to lose subscribers or affected its news syndication business.

In view of these factors, the Court declined interim relief, holding that the balance of convenience did not favour ANI and that it had failed to establish irreparable injury

The court, also prima facie, held that OpenAI's use of ANI's literary works for training the LLMs underlying ChatGPT was protected as fair dealing under Section 52(1)(a) of the Copyright Act and, therefore, did not amount to copyright infringement.

Separately, while dealing with ANI's reproduction claim, the court prima facie found that the examples relied upon by ANI did not establish substantial reproduction of its literary works or demonstrate memorisation or regurgitation of ANI's works by ChatGPT.