Microsoft Hit with Lawsuit over Pirated Books in AI Training—but Recent Court Wins Signal It Could Prevail

Microsoft faces a new trial of a group of eminent authors who accuse the Company of using pirated copies of their books to form its Megatron artificial intelligence model.

Placed Tuesday before a New York Federal Court, the trial is added to an increasing wave of legal challenges of writers, publishers and press organizations accusing the main developers of AI to exploit the material protected by copyright without consent.

The complainants – who include the American author of Prometheus Kai Bird, the essayist Jia Tolentino and the historian Daniel Okrent – had to train Microsoft in Megatron using a set of data containing nearly 200,000 pirated pounds, without informing or compensate for the authors. According to the complaint, the AI ​​model was designed not only to learn the language, but to imitate “the syntax, the voice and the themes” of the works of the authors, allowing it to generate text which is closely like their original expression.

“The Megatron model is not simply informed by our work,” says the complaint, “it is fundamentally built on it – without authorization and without compensation.”

The authors request an order from the court to block the alleged counterfeiting of Microsoft and statutory damages of up to $ 150,000 for each work, which could lead to millions of dollars in responsibility if the court settles in their favor. Microsoft did not immediately respond to requests for comments and the lawyer for the authors refused to talk about the case.

But while the trial seems to be part of a broader campaign of copyright holders to slow down the uncontrolled AI training, the recent decisions of the court suggest that Microsoft could actually have the upper hand.

During last week, two key decisions inclined in favor of AI companies. In one, the American district judge William Alsup judged that the use by anthropic of the books protected by copyright to form his system of Claude IA was under “fair use”, qualifying the process of “extremely transformative”. ALSUP has judged that the formation of a model to understand and generate a language based on large text strips was different from original works, thus satisfying one of the main tests for fair use under the American copyright law.

One day earlier, Judge Vince Chhabria rejected a similar case led against Meta on his Lama model, categorically declaring that there is no dispute of the transformative nature of what Meta had built. Chhabria also noted that the authors did not show that the AI ​​system had caused significant damage to the market for their books, another key factor to determine fair use.

Although none of the people applies directly to the Microsoft case, legal analysts claim that decisions could shape the way the judges assess similar complaints in the future. It is believed that the two judges indicated that the courts were leaned over towards the treatment of the training in AI as a transformative and fair use of the material protected by copyright, in particular when the exit does not compete directly or does not replace the original work.

Consecutive decisions are praised as provisional victories but important for companies like Anthropic, Meta, Openai and now potentially Microsoft. However, the trial of the authors takes a slightly different approach by stressing that the company would have used hacked copies of their books, which could complicate the defense of Microsoft even if the training itself is deemed legal.

The court may have to examine not only the legality of the AI ​​objective, but also the nature of the data it has been fed.

The situation highlights the complexities of the Act of Copyright at the age of AI. Rather than offering clarity, recent decisions have deepened the complexity of the issue. They suggest that although training on the material protected by copyright can, in many cases, be considered as fair use, there may still be legal consequences depending on the way in which this material has been acquired.

Now Microsoft’s case has become another test of the end of the limits of fair use and the start of the offense. Some analysts believe that this question will require the intervention of the Congress or the Supreme Court, or this evolving legal gray area should not remain unstable.