The U.S. Federal Court has finally approved a $1.5 billion copyright settlement between Anthropic and a group of authors and book publishers. The case, which revolved around AI training data sources, also led to the further dismantling of the issues of “reasonable use of model training” and “how to obtain training materials”.
Court approves $1.5 billion settlement
According to Reuters, judges of the Federal District Court for the Northern District of California signed a final warrant on Monday. This means that Anthropic can start paying settlement payments to authors and publishers who sue for copyright infringement.
This class action was granted preliminary approval last year. In the early stages of the case, the then judge William Alsup found that Anthropic had illegally downloaded and kept millions of copyrighted books.
Training is a reasonable use.
However, the court ' s judgement on the central legal issues of the case was not entirely reversed to the plaintiff. Alsup had previously supported a key claim by Anthropic that the use of copyright-protected text training AI models could be considered a reasonable use.
This judgement is considered to be an important node of the AI industry because it distinguishes “model training behaviour” from “training data acquisition”. That is to say, training in itself is not necessarily against the law, but the process of obtaining material may still constitute a separate violation or violation.
The source of piracy is the key to reconciliation.
Court documents show that the training library of Anthropic comes mainly from two sources: books purchased after they are scanned, and books downloaded from pirated websites such as Library Genesis and Pirate Library Mirror.
In the judge ' s view, the former approach could be established, and the latter practice itself would have legal implications. The section on pirated access to books could have gone into trial. Anthropic subsequently opted for a settlement to avoid the case continuing to the jury stage and the possibility of higher compensation.
There will be no industry-wide precedent.
Although the case was closed, it did not provide a final answer for the entire AI industry. The reason is that this “reasonable use” judgement comes from the District Court, and that when Anthropic chose to settle, the case would not be subject to appeal proceedings and would make it difficult to set a generally binding precedent.
Companies such as Google, Meta, Midjourney and OpenAI still face similar copyright litigation. Just last week, a group of publishers and authors, including Hachette, Cengage, Elsevier and the writer Scott Turow, also filed a class action against Google for training in copyright-protected works, Gemini.