The Algorithm And The Author

By Lalltaksh Joshi[1]

(AI, Copyright, And The Cases That Will Define A Generation)

October 1, 2026

The Bar Bulletin

Does a machine ‘steal’ or ‘study’ when it reads millions of books over hours to learn how to write? Where does inspiration snuff and infringement begin when GenAI produces music indistinguishable from a living artist’s catalogue, perhaps trained on those very recordings?

Can a machine’s output belong to anyone at all? Is it original work or theft when AI systems trawl through scientists’ private conversations to crack century-old mathematical problems?

These questions are no longer hypothetical. They are live before courts across the world, and our country and the answers are going to redraw the economics of every creative industry on Earth.

One of the consequential unresolved case is The New York Times v. OpenAI. Times filed exhibits showing ChatGPT reproduced near-verbatim Times passages.[2] The U.S. District Judge ordered OpenAI to produce 20 million chat logs[3]. Objection by OpenAI to the direction was unsuccessful[4]. Finality of such a direction is going to have far-reaching impact on the ‘liability’ versus ‘fair-use’ debate. The enormous data may be fairly released or it may not. If it’s honestly done, the data will reveal if AI’s training (and resulting output) violates copyrighted journalism. While a final ruling is pending, but in September, 2026, the Department of Justice, United States of America intervened and supported OpenAI contending that the creative possibilities and public benefits of training large language models on such content far outweigh any competitive harm[5].

Interestingly, whether the functions performed by OpenAI through ChatGPT serve the overall public interest, was directly considered and opined by the High Court of Delhi in landmark judgment dated 24^th July, 2026 in ANI’s suit against OpenAI[6]. The Delhi High Court first provided a meticulous account of how large language models function. The High Court held that AI training falls within Section 52(1)(a)(i) of the Copyright Act which excludes fair dealing for private or personal use, including research, from the purview of infringement. The High Court extended the concept of “research” through the doctrine of updating construction to encompass machine learning. Since training does not place copyrighted material before the public, the High Court characterised it as purely private use. ANI has appealed[7] and the Division Bench of the Delhi High Court issued notice to OpenAI. The appeal is pending and its decision is going to have a far-reaching impact.

Going back to the United States of America, two district courts have supplied diverging answers. In Bartz v. Anthropic, the Court held that AI training based on lawfully acquired books constitutes fair use while using pirated copies does not.[8] Once a copyrighted work is purchased then there is nothing wrong to read, re-read and recall it from memory. In July, 2026, the case settled for USD 1.5 billion; approximately USD 3,000 per work across around 482,000 registered works[9]. In Thomson Reuters v. Ross Intelligence, the Court rejected fair use by AI and found infringement of 2,243 legal headnotes[10]. The Court found that Ross’s use is not transformative. It used headnotes of Thomson Reuters as AI data to create non-generative AI legal research tool to compete with Westlaw itself. On 29^th September 2026, the U.S. Court of Appeals for the Third Circuit affirmed the decision of the U.S. District Court rejecting Ross’s fair-use defense [11].

In case of any copyrighted work, illegal methods of procurement are going to attract consequences under the applicable laws. The source of acquisition of copyrighted work was also examined in the above cases. In both Bartz and ANI, unlawfully acquired copyrighted work was not given protection of fair use and fair dealing respectively. Commercial incentive in a creative work lays the foundation of copyright law. Therefore, ‘fair dealing’ / ‘fair use’ defence cannot be permitted for training of AI systems upon unlawfully procured copyright work.

Importantly, these decisions were based on an examination of the alleged infringing works placed before the Court. There was an objective inquiry into what use was made using AI and comparing the generated and copyrighted work. This is the reason for upholding fair use in Bartz and ANI and rejecting fairness in use in Thomson Reuters. Applying the R.G. Anand test, the Delhi High Court in ANI case, found no substantial similarity between ANI’s news articles and ChatGPT’s responses. The test of substantial similarity between copyright and derivative work goes beyond the disputes raised by carriers of news against AI platforms and apply to concerns voiced by authors, singers, publishers, alike, which is why the 24^th July 2026 judgment of the Delhi High Court is a primer.

A creative output generated with the assistance of AI using prompts by an individual is not the copyrighted work of the AI platform but capable of copyright in the name of such individual. India’s Copyright Office delivered a separately instructive ruling in August 2026, rejecting Stephen Thaler’s application to register the AI system DABUS as author of an artistic work.[12] The work crossed the originality threshold under the Copyright Act but an author is the person who causes the work to be made. DABUS could not be that person. Thaler declined to re-register himself as author, and no valid copyright subsisted. The ruling aligns with the D.C. Circuit’s decision in Thaler v. Perlmutter[13]. A consensus appears to be that AI cannot be an author, but the human who directs and curates creation can be. Originality and authorship are separate inquiries — a distinction that opens a practical route for AI-assisted output to attract protection, provided human creative judgment is demonstrably in the chain.

Music adds its own texture. Universal Music Group, Concord, and ABKCO filed two suits against Anthropic covering more than 20,000 compositions and seeking damages exceeding USD 3 billion. Warner and UMG have settled with AI music platforms Suno and Udio; Sony continues to litigate. From Munich, on July 31, 2026[14], came arguably the first judicial ruling specifically targeting generative AI music. In GEMA v. Suno, the Munich Regional Court found memorization and reproduction of six protected music compositions as infringement. The Court compared the copyrighted work with outputs produced by Suno to arrive at its decision of similarity and violation of copyright.

The newest dimension is agentic AI which is a system that reason, investigate, and synthesise at scale. In September 2026, OpenAI announced a partial solution to the Navier-Stokes equations[15] achieved through a multi-agent system over 88 hours, 130 billion tokens, and more than USD 10 million. Mathematicians at NYU and within Anthropic allege the system accessed notes they had shared privately with AI tools and repurposed their approaches without attribution. OpenAI acknowledged it cannot rule out that de-identified usage data contributed to the result[16]. No court has addressed the copyright status of reasoning shared in confidence with an AI tool and later harvested by it’s developer. OpenAI has now formed an advisory group on the use of AI tools in mathematical research and training. The law, which already struggles to keep pace with generative AI, has not begun to catch up with agentic AI.

AI and copyright is neither a pro-AI nor a pro-rights holder question. It is case-by-case, output-dependent, and jurisdiction-specific. Courts are deciding it one filing at a time; markets are settling it one licensing agreement at a time.

Copyright law was designed to balance creator incentives against public access. How it answers them will determine not only who profits from the AI revolution, but whether creators continue to produce the work that makes the revolution possible.

[1]Lalltaksh Joshi, Advocate

[2]NYT v. OpenAI & Microsoft: Case Status May 2026

[3]OpenAI Must Turn Over Millions of Chat Logs in Copyright Litigation, NY Federal Judge Rules | Law.com

[4]NYT v. OpenAI & Microsoft: Case Status May 2026

[5]Trump administration backs OpenAI in New York Times’ copyright case over training of chatbots – The Hindu

[6]COR1_ABL24072026SC10282024_191123.pdf

[7]OpenAI vs ANI: Delhi HC refuses any ex-parte order | India News

[8]Bartz-v.-Anthropic-Order.pdf

[9]Harry Potter publisher to receive millions in Anthropic copyright settlement | AI (artificial intelligence) | The Guardian

[10]Thomson Reuters v. Ross — AI training-data lawsuit · fiund

[11]https://www.reuters.com/business/media-telecom/us-appeals-court-upholds-thomson-reuters-landmark-win-ai-training-lawsuit-2026-09-29/

[12] Order of Registrar Unnat P. Pandit, Copyright Office of India, dated August 31, 2026.