A federal judge in San Francisco reigned The formation of an AI model on works protected by copyright without the specific authorization to do so was not a violation of the law on copyright.
US District Judge William Alsup said that the Société d’Ia Anthropic could affirm a defense “for fair use” against allegations of copyright to train his Claude AI models on books protected by copyright. But the judge also judged that it was exactly how these books had been obtained.
Alsup supported anthropic’s assertion that he was “fair use” to buy millions of books, then digitize them for use in AI formation. The judge said it was not correct, however, for Anthropic to also have downloaded millions of pirated copies of books on the Internet, then maintained a digital library of these hacked copies.
The judge ordered a separate trial on the storage of anthropic of these hacked books, which could determine the responsibility of the company and any damage linked to this potential violation. The judge has not yet decided to grant the opportunity to grant collective appeal status, which could considerably increase the financial risks to Anthropic if it has proven to have violated the rights of the authors.
To note that it was a “fair use” for Anthropic to train his models of AI on books written by three authors – and Bartz, Charles Graeber and Kirk Wallace Johnson – who had filed a complaint against AI Society for copyright violations, Alsup answered a question that signed since the GENERATIVE BOOM data can be used for OPENAI. Models without the owner’s consent?
Dozens of proceedings linked to AI-et-Copyright have been filed in the past three years, most of which depend on the concept of fair use, a doctrine which allows the use of material protected by copyright without authorization if the use is sufficiently transforming-which must serve a new objective or add a new meaning, rather than simply copy or substitute the original work.
Alsup’s decision can establish a precedent for these other copyright cases – although it is also likely that many of these decisions will be called, which means that it will take years until there is clarity around AI and copyright in the United States
According to the judge’s decision, the use of anthropic of books to train Claude was “extremely transformative” and constituted “an equitable use under article 107 of the law on copyright”. Anthropic told court that his training on AI was not only authorized, but aligned with the spirit of the American copyright law, which it has argued “not only allows, but encourages” such use because it promotes human creativity. The company said that it had copied books to “study the writing of the complainants, extract non -conventing information from it and use what it has learned to create revolutionary technology”.
Although the training of AI models with data protected by copyright can be considered equitable use, the distinct action of Anthropic to build and store a consultable repository of pirated books is not, judged Alsup. Alsup noted that the fact that Anthropic had later bought a copy of a book which she previously stolen internet “will not denounce him of responsibility for theft, but this can affect the extent of legal damages”.
The judge also looked at Asking on the recognition of anthropic that he had turned to downloading of hacked books in order to save time and money in the construction of his AI models. “This prescription doubts that any accused incapacity can ever respond to its burden of explaining why the download of source copies from pirate sites that it could have purchased or otherwise accessible was itself reasonably necessary for any subsequent fair use,” said Alsup.
The “transformative” nature of AI outputs is important, but it is not the only thing that matters with regard to fair use. There are three other factors to consider: what type of work it is (creative works get more protection than factual); How much work is used (less, the better it is); And if the new use harms the market for the original.
For example, there is the undergoing case against Meta And Openai of the actress Sarah Silverman and two other authors, who filed prosecution against copyright in 2023, alleging that hacked versions of their works were used without the authorization to form models of AI language. Defendants recently argued The fact that the use is of the doctrine of fair use because the “study” of AI systems works to “learn” and create new transformer content.
The judge of the federal district Vince Chhabria stressed that even if this is true, AI systems are “changing considerably, you could even say erase, the labor market of this person”. But he also challenged the complainants, claiming that their lawyers had not provided enough evidence of the potential market impacts.
Alsup’s decision differs considerably from that of Chhabria on this point. Alsup said that if it was undoubtedly true that Claude could lead to increased competition for the works of the authors, this type of “competitive or creative displacement is not the type of competitive or creative displacement which concerns the law on copyright”. The purpose of copyright was to encourage the creation of new works, not to protect the authors of the competition, said Alsup, and he compared the authors’ objections to Claude to the fear that teaching schoolchildren to write well could also lead to an explosion of competing books.
Alsup also took note of his decision that Anthropic had built “railings” in Claude who were intended to prevent him from producing outings that directly plagued the books on which he had been formed.
Neither the anthropogenic lawyers nor the lawyers of the complainants immediately responded to the requests for comments on the decision of Alsup.
