Meta and Anthropic score legal victory over publishers in AI fight.

In the past week, two major artificial intelligence (AI) companies—Meta and Anthropic—secured early legal victories in cases brought by groups of authors over the training of their large language models (LLMs) on copyrighted material. However, these wins are not as straightforward as they might seem and have raised numerous questions about the future of AI and copyright law. Federal judges in the Northern District of California, William Alsup and Vince Chhabria, presided over these cases. In the Anthropic case, Judge Alsup ruled that the company's use of authors' books to train its AI model Claude constituted fair use, emphasizing the transformative nature of the AI. Notably, Anthropic had initially acquired books through piracy, which the judge heavily criticized. Although Anthropic later attempted to rectify this by legally purchasing books, removing their covers, and scanning them, the ruling indicated that the initial piracy could still lead to significant financial liabilities. Similarly, Judge Chhabria dismissed the authors' complaint against Meta for training its LLM Llama on their copyrighted material. Chhabria acknowledged the transformative aspect of the AI but was skeptical of its overall market impact. He pointed out that the Llama model was unlikely to produce more than 50 words from any given work, which he deemed insufficient to infringe copyright. Despite the dismissal, Chhabria's ruling suggested deep concerns about the potential harms of AI models, particularly their ability to flood the market with low-quality, derivative works that could undermine the incentives for human creators. Both judges focused primarily on the training phase of the LLMs, rather than the outputs. They highlighted that the transformation of input material into something new was a key factor in their decisions. However, the output of AI models remains a critical unresolved issue. For instance, The New York Times and Disney have filed lawsuits against OpenAI and Midjourney, respectively, over claims that their AI models can reproduce copyrighted content, such as Times articles and Disney characters. These cases underscore the complexity and ambiguity surrounding AI-generated content and its potential legal ramifications. The legal strategies employed by the authors in both cases were deemed inadequate by the judges. Chhabria explicitly stated that his ruling did not mean Meta's use of copyrighted materials was lawful, but rather that the plaintiffs made the wrong arguments. He suggested that future lawsuits could present stronger cases, particularly if they address the broader market impact of AI outputs. Alsup, in his ruling, compared the authors' complaints to objections against training schoolchildren to write, arguing that the Copyright Act aims to promote original works, not shield authors from competition. These rulings have set a preliminary legal framework for AI training, but they have left many questions open, particularly regarding the output of AI models. If AI outputs were found to infringe copyright, the legal landscape could shift dramatically. The current generation of AI models, such as Claude and Llama, are essentially useless without generating outputs, making this an urgent issue to resolve. Industry insiders and legal experts are divided on the implications of these rulings. Some argue that requiring permission for training data could stifle innovation, particularly in the open-source community, where access to proprietary data is limited. Former Meta executive Nick Clegg went so far as to claim that obtaining such permissions would "basically kill the AI industry." However, given the increasing number of licensing deals being struck, this view is becoming less tenable. Many large AI companies, backed by significant investments, can afford to comply with these requirements, although smaller players might struggle. Others, like legal expert Blake Reid, highlight the potential for massive financial damages if evidence of piracy is unearthed. Anthropic's and Meta's experiences suggest that companies previously involved in piracy could face severe financial penalties. Reid vividly described the situation: "If there’s evidence that an engineer was torrenting a bunch of stuff with C-suite blessing, it turns the company into a money piñata." Looking ahead, these rulings could encourage more rigorous scrutiny of how AI models are trained. Companies might be forced to invest more in legally acquiring data, which could slow down development but also lead to more ethical practices. However, the broader impact on the creative industries remains uncertain. If artists are compensated for their work being used in training data, it could provide a short-term financial boost. But the long-term consequences—such as the market being flooded with AI-generated content—might be detrimental to new and emerging artists. The courts, while beginning to weigh in, are still grappling with the nuances of AI and its implications for copyright law. These early decisions are important but far from conclusive. As the legal battles continue, the industry will need to navigate a complex and evolving landscape, balancing technological innovation with the protection of creative rights. Vox Media, the parent company of The Verge, has a technology and content deal with OpenAI, adding a layer of interest to The Verge’s coverage of these events. The dual rulings represent a pivotal moment in the ongoing debate over AI and intellectual property, but they also highlight the need for clearer, more comprehensive legal guidelines and a deeper understanding of the ethical and economic implications of AI in creative fields.