Subscribe
July was packed with cases pivoting on the legality of artificial intelligence (AI) companies helping themselves to data across the world to train their large language models. The lawsuits involved the critical issue of copyright protection, in particular for publishing companies and a range of creative fields such as literature, cinema and music, as AI behemoths such as OpenAI, Google, Anthropic and Meta freely mine data to train their machines. Is it copyright infringement, as courts in Europe have held, or is it fair use as the US courts have ruled, albeit in rather confusing verdicts?
The Delhi High Court also weighed in on this critical question in a case filed by news agency ANI (Asian News International) against OpenAI and said in a preliminary ruling that the AI entity’s use of copyright material did not constitute infringement but was fair use, or “fair dealing” under India’s copyright law. This has raised vexed questions on how far the law is applicable for AI operations since it does not specify any appropriate exceptions for this industry.
The copyright law and its interpretation here and in the US and the UK in relation to AI are fluid, unlike in the EU, which has a clearly defined AI Act brought in 2024. This codifies new requirements for AI model training under the copyright law and stipulates that reproductions made for AI training must be authorised by rights holders, unless covered by existing copyright exceptions such as text and data mining (TDM) provisions.
Thus, a regional court in Munich had no difficulty in ruling against OpenAI in a case filed by German music copyright organisation GEMA. It stated that the use of copyrighted song lyrics for training generative AI models without a licence violated German copyright law. The case centred on lyrics from well-known German songs which were being generated by Open AI’s chatbot ChatGPT through simple prompts. The court rejected Open AI’s untenable argument that users, and not the platform, were responsible for any infringement. The Munich court also dismissed reliance on innovation-driven copyright exceptions. This is the first judgement that provides clarity on copyright infringement by AI models and is likely to reshape AI compliance across Europe.
What is happening in the US, however, is confusing to say the least. Towards the end of July, a US federal judge gave final approval to a record US $1.5 billion class action settlement in a suit filed in 2024 by a large group of authors against Anthropic over the misuse of their books to train its AI chatbot Claude. They accused Anthropic, the biggest AI firm with a valuation of $965 billion of retaining and storing their works, totalling nearly half a million pirated books, without authorisation and sued for damages. Here’s how bizarre the ruling is. The court said that AI model training could copy the books without being liable for copyright infringement but storing pirated, shadow-library text exposed the company to legal challenges. What does this amount to? Downloading an unauthorised e-book creates liability, whereas buying the same book and physically dismantling it qualifies as fair use!
Although the $1.5 settlement appears humongous and is one for the record books, it translates to very little for individual authors after the lawyers’ fees are paid. Charles Graeber, author of the bestselling The Good Nurse, who led the fight with a clutch of fellow writers, revealed that the victory came at a steep cost to himself. He got just $6,200 for the use of two unauthorised books.
The confrontation of rights- holders with AI companies over intellectual property rights, which started in 2024, is intensifying. In May, a group of publishers, accounting for the biggest names in the industry, joined forces to take on Meta and its head Mark Zuckerberg over the unauthorised use of their copyrighted books and academic materials to train Llama, the company’s LLM. This time the legal battle is evenly matched because all the heavyweights of publishing are involved in the class action suit filed in a federal court in New York. Reports have termed it one of the most significant copyright challenges brought against a major AI developer since the plaintiffs include Elsevier, Hachette Book Group, Macmillan Publishers, McGraw Hill and Cengage Learning.
Their complaint is that Meta allegedly reproduced and distributed “millions of copyrighted works” without authorisation or compensation in order to develop and improve its AI systems. They also accuse Zuckerberg of personally approving and encouraging the practices, claiming that Meta subscribed to a “move fast and break things” approach when dealing with copyrighted material. Meta has rejected the allegations maintaining that AI training qualifies as fair use under US copyright law. There is also the customary justification that AI is driving innovation, productivity and creativity across industries.
This is an argument that Justice Amit Bansal of the Delhi High Court appears to have bought into. Refusing ANI's plea for an interim injunction, Justice Amit Bansal said that using copyrighted material to train LLMs falls within the fair dealing exemption for research under the Copyright Act, 1957. The court also believed that “irreparable injury would be caused not only to OpenAI but also to the public at large” if an interim injunction was granted in ANI’s favour. Now, that begs several questions. Is the court deciding on the injury caused to the plaintiff or assuming unsubstantiated damage to the megacorporation, which is valued at $852 billion and generates roughly $2 billion in monthly revenue?
ANI had filed a lawsuit against OpenAI in the Delhi High Court in November 2024, alleging that the US-based company used its published content, some of it behind paywalls, without permission to train its AI models and that ChatGPT attributed fabricated stories to the news agency. ANI has alleged that ChatGPT reproduced certain content from its proprietary journalism whenever it was prompted. It, therefore, sought monetary damages of Rs 2 crore and a permanent injunction against OpenAI to stop using any additional content.
A number of stakeholders including the Digital News Publishers Association, the Indian Music Industry, the Federation of Indian Publishers, and technology-based start-ups have filed interventions in this suit, turning the case into a major policy question on what can be termed fair use by the AI industry. The outcome will have implications for every creative field from cinema to music, for every publisher and for the future of media since AI systems train on their output without consent or compensation, and then build a commercial product to compete against them.
